{
  "schema_version": "generalgamebench-snapshot/1",
  "season": "hosted-rerun-2026-10-05",
  "generated_at": "2026-10-05T21:50:31.662849+00:00",
  "exhibition": [
    {
      "agent": "openai-gpt-6-astra",
      "mode": "exhibition",
      "hardware": "Darwin arm64",
      "version": "0.5.0",
      "max_steps": 8,
      "score": 18.125,
      "ci95": null,
      "episodes": 43,
      "seeds": 1,
      "seed_ids": [
        3000
      ],
      "games": [
        "coin-run",
        "dodge-lanes",
        "doom-basic",
        "doom-defend-center",
        "doom-defend-line",
        "doom-take-cover",
        "doom-health",
        "doom-corridor",
        "doom-home",
        "doom-predict",
        "procgen-bigfish",
        "procgen-bossfight",
        "procgen-caveflyer",
        "procgen-chaser",
        "procgen-climber",
        "procgen-coinrun",
        "procgen-dodgeball",
        "procgen-fruitbot",
        "procgen-heist",
        "procgen-jumper",
        "procgen-leaper",
        "procgen-maze",
        "procgen-miner",
        "procgen-ninja",
        "procgen-plunder",
        "procgen-starpilot",
        "crafter",
        "miniworld-oneroom",
        "pettingzoo-pistonball",
        "atari-breakout",
        "retro-airstriker",
        "nethack-score",
        "minihack-room",
        "unity-food-collector",
        "football-empty-goal",
        "supertuxkart-lighthouse",
        "luanti-chop-tree",
        "supertux-first-coin",
        "dcss-first-experience",
        "openttd-first-road",
        "mindustry-copper",
        "cdda-first-weapon",
        "warzone-first-derrick"
      ],
      "task_metadata": {
        "coin-run": {},
        "dodge-lanes": {},
        "doom-basic": {},
        "doom-defend-center": {},
        "doom-defend-line": {},
        "doom-take-cover": {},
        "doom-health": {},
        "doom-corridor": {},
        "doom-home": {},
        "doom-predict": {},
        "procgen-bigfish": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            40
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bigfish",
          "task_version": "1"
        },
        "procgen-bossfight": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bossfight",
          "task_version": "1"
        },
        "procgen-caveflyer": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2,
            12
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-caveflyer",
          "task_version": "1"
        },
        "procgen-chaser": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-chaser",
          "task_version": "1"
        },
        "procgen-climber": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            12.6
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-climber",
          "task_version": "1"
        },
        "procgen-coinrun": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-coinrun",
          "task_version": "1"
        },
        "procgen-dodgeball": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            19
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-dodgeball",
          "task_version": "1"
        },
        "procgen-fruitbot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -1.5,
            32.4
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-fruitbot",
          "task_version": "1"
        },
        "procgen-heist": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-heist",
          "task_version": "1"
        },
        "procgen-jumper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-jumper",
          "task_version": "1"
        },
        "procgen-leaper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-leaper",
          "task_version": "1"
        },
        "procgen-maze": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-maze",
          "task_version": "1"
        },
        "procgen-miner": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-miner",
          "task_version": "1"
        },
        "procgen-ninja": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-ninja",
          "task_version": "1"
        },
        "procgen-plunder": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            4.5,
            30
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-plunder",
          "task_version": "1"
        },
        "procgen-starpilot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2.5,
            64
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-starpilot",
          "task_version": "1"
        },
        "crafter": {
          "engine": "crafter",
          "engine_version": "1.8.3",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            22
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "crafter",
          "task_version": "2"
        },
        "miniworld-oneroom": {
          "engine": "miniworld",
          "engine_version": "2.1.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "miniworld-oneroom",
          "task_version": "1"
        },
        "pettingzoo-pistonball": {
          "engine": "pettingzoo",
          "engine_version": "1.27.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -100,
            100
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "pettingzoo-pistonball",
          "task_version": "1"
        },
        "atari-breakout": {
          "engine": "ale-py",
          "engine_version": "0.12.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            864
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "atari-breakout",
          "task_version": "1"
        },
        "retro-airstriker": {
          "engine": "stable-retro",
          "engine_version": "1.0.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            10000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "retro-airstriker",
          "task_version": "1"
        },
        "nethack-score": {
          "engine": "nle",
          "engine_version": "1.3.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "nethack-score",
          "task_version": "1"
        },
        "minihack-room": {
          "engine": "minihack",
          "engine_version": "1.0.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "minihack-room",
          "task_version": "1"
        },
        "unity-food-collector": {
          "assets_sha256": "c28e27b2d3af4391a3e1e95b0871e01a16ef657ee062bb91e7740ac9d0c99be5",
          "camera_size": [
            84,
            84
          ],
          "capture_frame_rate": 30,
          "controlled_agent": "lowest-initial-agent-id",
          "engine": "mlagents-envs",
          "engine_version": "1.1.0",
          "numpy_version": "1.23.5",
          "observation": "rendered-rgb-only",
          "peer_policy": "no-op",
          "replay": "exact-frame-and-score",
          "reward_definition": "sum of native food rewards; green +1, red -1",
          "scene": "VisualFoodCollector",
          "score_range": [
            0,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "unity-food-collector",
          "task_version": "1"
        },
        "football-empty-goal": {
          "camera_size": [
            320,
            180
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:8774faa312fa0791f01966ef3152d23c012dda5117fd01366972a0bebfc13635",
          "control_policy": "release movement and shot before each decision",
          "engine": "gfootball",
          "engine_revision": "ba9952130898bde2879b01ff2b49d3d48d26ce81",
          "engine_version": "2.10.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native goal reward, no checkpoint shaping",
          "scenario": "academy_empty_goal_close",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "football-empty-goal",
          "task_version": "1"
        },
        "supertuxkart-lighthouse": {
          "assets_sha256": "56cd754653224efde4a40aef0400a71f03f72f18d2c35a89896db052dc096b23",
          "camera_size": [
            320,
            240
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:257166494f158e6b5b3441d4b52f99b83e8cef4ff60c6ce5b333251aa7cf4738",
          "engine": "PySuperTuxKart2",
          "engine_revision": "dd70f6823f248ae1df2ce513839a9b2c8c940c39",
          "engine_variant": "deterministic-render-v2",
          "engine_version": "0.7.4+ggbrender2",
          "neutral_preroll_steps": 50,
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "Mesa software / deterministic-render-v2",
          "renderer_threads": 1,
          "renderer_vector_bits": 128,
          "replay": "exact-frame-and-score",
          "reward_definition": "max(0, native overall distance) / native track length",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_seconds": 0.2,
          "task_id": "supertuxkart-lighthouse",
          "task_version": "3",
          "track": "lighthouse"
        },
        "luanti-chop-tree": {
          "camera_size": [
            64,
            64
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:f3f2df899b8709a904350472825f3253fd1a46a8d913322634a604e389fbb0fe",
          "engine": "craftium",
          "engine_revision": "8cffe4176e793f78d00b17fa7e46ccf333a7b5b0",
          "engine_variant": "serial-lockstep-v2",
          "engine_version": "0.0.1",
          "mesh_generation_threads": 1,
          "numpy_version": "2.5.3",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native Lua tree-node dig reward; first reward completes task",
          "scenario": "Craftium/ChopTree-v0: first tree block",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "pinned world snapshot; fixed map/Lua seed on initial reset",
          "simulation": "lockstep",
          "step_seconds": 0.02,
          "task_id": "luanti-chop-tree",
          "task_version": "1",
          "weather": "disabled via upstream enable_weather setting"
        },
        "supertux-first-coin": {
          "camera_size": [
            640,
            480
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:b3f936e8da2a38c0eab38ea1acfe56c1c47829cbe1f288775a135f0a67faadbf",
          "engine": "SuperTux",
          "engine_revision": "c1ddb4f28c54de77f07aa9965231a91c0ed06708",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.6.3+ggbench1",
          "frames_per_action": 8,
          "initialization": "skip level-introduction menu and fades; fresh player/save",
          "level": "world1/welcome_antarctica.stl",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "physics_step_seconds": 0.015,
          "renderer": "SDL software",
          "replay": "exact-frame-and-score",
          "reward_definition": "native level coin count; first coin completes task",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "supertux-first-coin",
          "task_version": "1"
        },
        "dcss-first-experience": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:942c43c70e0682b3d612e7c0b6e42eb8b35aeef42aad8bb77982bbbad4d617b8",
          "control_policy": "one native key per decision; no auto-explore or auto-fight",
          "engine": "DCSS",
          "engine_revision": "19cb0921cbcb6463fcedd1bb06636446c50b1905",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.34.0+ggbench1",
          "initialization": "advance title and seed-confirmation menus before observing",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 tiles / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native experience gain; death ends task",
          "scenario": "Dungeon:1; Minotaur Fighter with war axe; first experience",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "native uint64 seed; benchmark zero maps to uint64 maximum",
          "simulation": "lockstep",
          "task_id": "dcss-first-experience",
          "task_version": "1"
        },
        "openttd-first-road": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:227f7a13d0f6a96f2574ac0b428755101ad54def2217d629d08d059f4e2996b0",
          "cursor_step_pixels": 32,
          "engine": "OpenTTD",
          "engine_revision": "14ec60f248547d4d062a1160f0fc26d742319888",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "15.3+ggbench1",
          "initialization": "16 native ticks; fresh company and configuration",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native 32bpp software framebuffer; OpenGFX 7.1",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when native company infrastructure owns at least two road bits",
          "scenario": "new 64x64 company in 1950; first owned road",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 30,
          "task_id": "openttd-first-road",
          "task_version": "1"
        },
        "mindustry-copper": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:90468c4e5dde3dfaf18e1ec5fe337b4d32906271d3a7be899efd5b87756082a2",
          "cursor_step_pixels": 32,
          "engine": "Mindustry",
          "engine_revision": "067c720a8817c1c9fb586c03898a7d948caaed56",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "160.5+ggbench1",
          "frames_per_action": 6,
          "initialization": "120 native frames; fresh player and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "optional_effects": false,
          "physics_step_seconds": 0.016666666666666666,
          "renderer": "native SDL/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "native player-plus-core copper gain from initial total; target 15",
          "scenario": "serpulo/groundZero; native first-15-copper tutorial objective",
          "score_range": [
            0,
            15
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "mindustry-copper",
          "task_version": "1"
        },
        "cdda-first-weapon": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:eee69b35296e92b6ddc84a9ffbf737ab66ab48ae42dbbc3f1ee8e236b97c1b1f",
          "control_policy": "one native key per decision; tutorial popups remain visible",
          "engine": "Cataclysm: Dark Days Ahead",
          "engine_revision": "7b2efa5cea38e4d4d97dd0e63b28b9148623da59",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.I-1+ggbench1",
          "initialization": "native tutorial world and preset character; fresh user directory",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 software / UltimateCataclysm tiles",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when the native avatar wields the tutorial baseball bat",
          "scenario": "native tutorial; first wielded baseball bat",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "cdda-first-weapon",
          "task_version": "1"
        },
        "warzone-first-derrick": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:c2c0aee72affd4e5d5be6aa03659c533e3c968365e0f62eccf1f9fbdd30f6c22",
          "cursor_step_pixels": 32,
          "engine": "Warzone 2100",
          "engine_revision": "7d3d8e51a61ceee89e6bf672a41337e1a9cd7392",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "4.7.0+ggbench1",
          "fine_cursor_step_pixels": 8,
          "initialization": "40 native frames; fresh tutorial and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL3/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native completed structure; two unit losses end task",
          "scenario": "native TUTORIAL3; first constructed oil derrick",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 100,
          "task_id": "warzone-first-derrick",
          "task_version": "1"
        }
      },
      "decisions": 342,
      "p50_ms": 9661.910583500001,
      "p95_ms": 18023.86950655,
      "max_ms": 22389.972,
      "misses": 342,
      "latency_eligible": false,
      "latency_policy": {
        "id": "suite-p95-200-v1",
        "metric": "suite_p95_ms",
        "limit_ms": 200.0,
        "comparison": "strictly_less_than",
        "quantile_method": "linear",
        "sample_scope": "all_recorded_decisions_in_complete_suite"
      },
      "response_timeout_seconds": 120.0,
      "trust": "local-unattested",
      "official_rank": null,
      "per_game": {
        "coin-run": 100.0,
        "dodge-lanes": 100.0,
        "doom-basic": 66.19,
        "doom-defend-center": 9.524,
        "doom-defend-line": 14.286,
        "doom-take-cover": 100.0,
        "doom-health": 100.0,
        "doom-corridor": 9.091,
        "doom-home": 0.0,
        "doom-predict": 76.183,
        "procgen-bigfish": 0.0,
        "procgen-bossfight": 0.0,
        "procgen-caveflyer": 0.0,
        "procgen-chaser": 0.0,
        "procgen-climber": 0.0,
        "procgen-coinrun": 0.0,
        "procgen-dodgeball": 0.0,
        "procgen-fruitbot": 4.425,
        "procgen-heist": 0.0,
        "procgen-jumper": 0.0,
        "procgen-leaper": 0.0,
        "procgen-maze": 0.0,
        "procgen-miner": 0.0,
        "procgen-ninja": 0.0,
        "procgen-plunder": 0.0,
        "procgen-starpilot": 0.0,
        "crafter": 4.545,
        "miniworld-oneroom": 0.0,
        "pettingzoo-pistonball": 93.775,
        "atari-breakout": 0.0,
        "retro-airstriker": 0.0,
        "nethack-score": 0.0,
        "minihack-room": 100.0,
        "unity-food-collector": 0.0,
        "football-empty-goal": 0.0,
        "supertuxkart-lighthouse": 1.362,
        "luanti-chop-tree": 0.0,
        "supertux-first-coin": 0.0,
        "dcss-first-experience": 0.0,
        "openttd-first-road": 0.0,
        "mindustry-copper": 0.0,
        "cdda-first-weapon": 0.0,
        "warzone-first-derrick": 0.0
      },
      "errors": 0,
      "aborted_episodes": 0,
      "model": "gpt-6-astra",
      "provider_metadata": {
        "transport": "authenticated-cli-per-frame",
        "requested_model": "gpt-6-astra",
        "tool_events": 0,
        "observation": "png-only",
        "calls": 342,
        "prompt_version": "pixels-json-action/2"
      },
      "hardware_details": "Apple M3 Max; 48 GiB RAM; 16 CPU cores; macOS arm64. One hosted model and one game run at a time. Each action starts a fresh authenticated CLI call. CLI startup, image transfer, network time, provider inference and action validation remain in the timer. Provider hardware is not known. Normal desktop tasks can use the same Mac."
    },
    {
      "agent": "claude-claude-opus-5-1m-",
      "mode": "exhibition",
      "hardware": "Darwin arm64",
      "version": "0.5.0",
      "max_steps": 8,
      "score": 17.998,
      "ci95": null,
      "episodes": 43,
      "seeds": 1,
      "seed_ids": [
        3000
      ],
      "games": [
        "coin-run",
        "dodge-lanes",
        "doom-basic",
        "doom-defend-center",
        "doom-defend-line",
        "doom-take-cover",
        "doom-health",
        "doom-corridor",
        "doom-home",
        "doom-predict",
        "procgen-bigfish",
        "procgen-bossfight",
        "procgen-caveflyer",
        "procgen-chaser",
        "procgen-climber",
        "procgen-coinrun",
        "procgen-dodgeball",
        "procgen-fruitbot",
        "procgen-heist",
        "procgen-jumper",
        "procgen-leaper",
        "procgen-maze",
        "procgen-miner",
        "procgen-ninja",
        "procgen-plunder",
        "procgen-starpilot",
        "crafter",
        "miniworld-oneroom",
        "pettingzoo-pistonball",
        "atari-breakout",
        "retro-airstriker",
        "nethack-score",
        "minihack-room",
        "unity-food-collector",
        "football-empty-goal",
        "supertuxkart-lighthouse",
        "luanti-chop-tree",
        "supertux-first-coin",
        "dcss-first-experience",
        "openttd-first-road",
        "mindustry-copper",
        "cdda-first-weapon",
        "warzone-first-derrick"
      ],
      "task_metadata": {
        "coin-run": {},
        "dodge-lanes": {},
        "doom-basic": {},
        "doom-defend-center": {},
        "doom-defend-line": {},
        "doom-take-cover": {},
        "doom-health": {},
        "doom-corridor": {},
        "doom-home": {},
        "doom-predict": {},
        "procgen-bigfish": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            40
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bigfish",
          "task_version": "1"
        },
        "procgen-bossfight": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bossfight",
          "task_version": "1"
        },
        "procgen-caveflyer": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2,
            12
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-caveflyer",
          "task_version": "1"
        },
        "procgen-chaser": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-chaser",
          "task_version": "1"
        },
        "procgen-climber": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            12.6
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-climber",
          "task_version": "1"
        },
        "procgen-coinrun": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-coinrun",
          "task_version": "1"
        },
        "procgen-dodgeball": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            19
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-dodgeball",
          "task_version": "1"
        },
        "procgen-fruitbot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -1.5,
            32.4
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-fruitbot",
          "task_version": "1"
        },
        "procgen-heist": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-heist",
          "task_version": "1"
        },
        "procgen-jumper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-jumper",
          "task_version": "1"
        },
        "procgen-leaper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-leaper",
          "task_version": "1"
        },
        "procgen-maze": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-maze",
          "task_version": "1"
        },
        "procgen-miner": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-miner",
          "task_version": "1"
        },
        "procgen-ninja": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-ninja",
          "task_version": "1"
        },
        "procgen-plunder": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            4.5,
            30
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-plunder",
          "task_version": "1"
        },
        "procgen-starpilot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2.5,
            64
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-starpilot",
          "task_version": "1"
        },
        "crafter": {
          "engine": "crafter",
          "engine_version": "1.8.3",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            22
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "crafter",
          "task_version": "2"
        },
        "miniworld-oneroom": {
          "engine": "miniworld",
          "engine_version": "2.1.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "miniworld-oneroom",
          "task_version": "1"
        },
        "pettingzoo-pistonball": {
          "engine": "pettingzoo",
          "engine_version": "1.27.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -100,
            100
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "pettingzoo-pistonball",
          "task_version": "1"
        },
        "atari-breakout": {
          "engine": "ale-py",
          "engine_version": "0.12.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            864
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "atari-breakout",
          "task_version": "1"
        },
        "retro-airstriker": {
          "engine": "stable-retro",
          "engine_version": "1.0.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            10000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "retro-airstriker",
          "task_version": "1"
        },
        "nethack-score": {
          "engine": "nle",
          "engine_version": "1.3.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "nethack-score",
          "task_version": "1"
        },
        "minihack-room": {
          "engine": "minihack",
          "engine_version": "1.0.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "minihack-room",
          "task_version": "1"
        },
        "unity-food-collector": {
          "assets_sha256": "c28e27b2d3af4391a3e1e95b0871e01a16ef657ee062bb91e7740ac9d0c99be5",
          "camera_size": [
            84,
            84
          ],
          "capture_frame_rate": 30,
          "controlled_agent": "lowest-initial-agent-id",
          "engine": "mlagents-envs",
          "engine_version": "1.1.0",
          "numpy_version": "1.23.5",
          "observation": "rendered-rgb-only",
          "peer_policy": "no-op",
          "replay": "exact-frame-and-score",
          "reward_definition": "sum of native food rewards; green +1, red -1",
          "scene": "VisualFoodCollector",
          "score_range": [
            0,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "unity-food-collector",
          "task_version": "1"
        },
        "football-empty-goal": {
          "camera_size": [
            320,
            180
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:8774faa312fa0791f01966ef3152d23c012dda5117fd01366972a0bebfc13635",
          "control_policy": "release movement and shot before each decision",
          "engine": "gfootball",
          "engine_revision": "ba9952130898bde2879b01ff2b49d3d48d26ce81",
          "engine_version": "2.10.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native goal reward, no checkpoint shaping",
          "scenario": "academy_empty_goal_close",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "football-empty-goal",
          "task_version": "1"
        },
        "supertuxkart-lighthouse": {
          "assets_sha256": "56cd754653224efde4a40aef0400a71f03f72f18d2c35a89896db052dc096b23",
          "camera_size": [
            320,
            240
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:257166494f158e6b5b3441d4b52f99b83e8cef4ff60c6ce5b333251aa7cf4738",
          "engine": "PySuperTuxKart2",
          "engine_revision": "dd70f6823f248ae1df2ce513839a9b2c8c940c39",
          "engine_variant": "deterministic-render-v2",
          "engine_version": "0.7.4+ggbrender2",
          "neutral_preroll_steps": 50,
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "Mesa software / deterministic-render-v2",
          "renderer_threads": 1,
          "renderer_vector_bits": 128,
          "replay": "exact-frame-and-score",
          "reward_definition": "max(0, native overall distance) / native track length",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_seconds": 0.2,
          "task_id": "supertuxkart-lighthouse",
          "task_version": "3",
          "track": "lighthouse"
        },
        "luanti-chop-tree": {
          "camera_size": [
            64,
            64
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:f3f2df899b8709a904350472825f3253fd1a46a8d913322634a604e389fbb0fe",
          "engine": "craftium",
          "engine_revision": "8cffe4176e793f78d00b17fa7e46ccf333a7b5b0",
          "engine_variant": "serial-lockstep-v2",
          "engine_version": "0.0.1",
          "mesh_generation_threads": 1,
          "numpy_version": "2.5.3",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native Lua tree-node dig reward; first reward completes task",
          "scenario": "Craftium/ChopTree-v0: first tree block",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "pinned world snapshot; fixed map/Lua seed on initial reset",
          "simulation": "lockstep",
          "step_seconds": 0.02,
          "task_id": "luanti-chop-tree",
          "task_version": "1",
          "weather": "disabled via upstream enable_weather setting"
        },
        "supertux-first-coin": {
          "camera_size": [
            640,
            480
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:b3f936e8da2a38c0eab38ea1acfe56c1c47829cbe1f288775a135f0a67faadbf",
          "engine": "SuperTux",
          "engine_revision": "c1ddb4f28c54de77f07aa9965231a91c0ed06708",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.6.3+ggbench1",
          "frames_per_action": 8,
          "initialization": "skip level-introduction menu and fades; fresh player/save",
          "level": "world1/welcome_antarctica.stl",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "physics_step_seconds": 0.015,
          "renderer": "SDL software",
          "replay": "exact-frame-and-score",
          "reward_definition": "native level coin count; first coin completes task",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "supertux-first-coin",
          "task_version": "1"
        },
        "dcss-first-experience": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:942c43c70e0682b3d612e7c0b6e42eb8b35aeef42aad8bb77982bbbad4d617b8",
          "control_policy": "one native key per decision; no auto-explore or auto-fight",
          "engine": "DCSS",
          "engine_revision": "19cb0921cbcb6463fcedd1bb06636446c50b1905",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.34.0+ggbench1",
          "initialization": "advance title and seed-confirmation menus before observing",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 tiles / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native experience gain; death ends task",
          "scenario": "Dungeon:1; Minotaur Fighter with war axe; first experience",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "native uint64 seed; benchmark zero maps to uint64 maximum",
          "simulation": "lockstep",
          "task_id": "dcss-first-experience",
          "task_version": "1"
        },
        "openttd-first-road": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:227f7a13d0f6a96f2574ac0b428755101ad54def2217d629d08d059f4e2996b0",
          "cursor_step_pixels": 32,
          "engine": "OpenTTD",
          "engine_revision": "14ec60f248547d4d062a1160f0fc26d742319888",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "15.3+ggbench1",
          "initialization": "16 native ticks; fresh company and configuration",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native 32bpp software framebuffer; OpenGFX 7.1",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when native company infrastructure owns at least two road bits",
          "scenario": "new 64x64 company in 1950; first owned road",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 30,
          "task_id": "openttd-first-road",
          "task_version": "1"
        },
        "mindustry-copper": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:90468c4e5dde3dfaf18e1ec5fe337b4d32906271d3a7be899efd5b87756082a2",
          "cursor_step_pixels": 32,
          "engine": "Mindustry",
          "engine_revision": "067c720a8817c1c9fb586c03898a7d948caaed56",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "160.5+ggbench1",
          "frames_per_action": 6,
          "initialization": "120 native frames; fresh player and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "optional_effects": false,
          "physics_step_seconds": 0.016666666666666666,
          "renderer": "native SDL/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "native player-plus-core copper gain from initial total; target 15",
          "scenario": "serpulo/groundZero; native first-15-copper tutorial objective",
          "score_range": [
            0,
            15
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "mindustry-copper",
          "task_version": "1"
        },
        "cdda-first-weapon": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:eee69b35296e92b6ddc84a9ffbf737ab66ab48ae42dbbc3f1ee8e236b97c1b1f",
          "control_policy": "one native key per decision; tutorial popups remain visible",
          "engine": "Cataclysm: Dark Days Ahead",
          "engine_revision": "7b2efa5cea38e4d4d97dd0e63b28b9148623da59",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.I-1+ggbench1",
          "initialization": "native tutorial world and preset character; fresh user directory",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 software / UltimateCataclysm tiles",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when the native avatar wields the tutorial baseball bat",
          "scenario": "native tutorial; first wielded baseball bat",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "cdda-first-weapon",
          "task_version": "1"
        },
        "warzone-first-derrick": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:c2c0aee72affd4e5d5be6aa03659c533e3c968365e0f62eccf1f9fbdd30f6c22",
          "cursor_step_pixels": 32,
          "engine": "Warzone 2100",
          "engine_revision": "7d3d8e51a61ceee89e6bf672a41337e1a9cd7392",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "4.7.0+ggbench1",
          "fine_cursor_step_pixels": 8,
          "initialization": "40 native frames; fresh tutorial and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL3/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native completed structure; two unit losses end task",
          "scenario": "native TUTORIAL3; first constructed oil derrick",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 100,
          "task_id": "warzone-first-derrick",
          "task_version": "1"
        }
      },
      "decisions": 340,
      "p50_ms": 2353.58125,
      "p95_ms": 3273.0903916500006,
      "max_ms": 9040.905333,
      "misses": 340,
      "latency_eligible": false,
      "latency_policy": {
        "id": "suite-p95-200-v1",
        "metric": "suite_p95_ms",
        "limit_ms": 200.0,
        "comparison": "strictly_less_than",
        "quantile_method": "linear",
        "sample_scope": "all_recorded_decisions_in_complete_suite"
      },
      "response_timeout_seconds": 120.0,
      "trust": "local-unattested",
      "official_rank": null,
      "per_game": {
        "coin-run": 100.0,
        "dodge-lanes": 100.0,
        "doom-basic": 66.19,
        "doom-defend-center": 4.762,
        "doom-defend-line": 9.524,
        "doom-take-cover": 100.0,
        "doom-health": 100.0,
        "doom-corridor": 9.091,
        "doom-home": 0.0,
        "doom-predict": 76.183,
        "procgen-bigfish": 0.0,
        "procgen-bossfight": 0.0,
        "procgen-caveflyer": 0.0,
        "procgen-chaser": 0.0,
        "procgen-climber": 0.0,
        "procgen-coinrun": 0.0,
        "procgen-dodgeball": 0.0,
        "procgen-fruitbot": 4.425,
        "procgen-heist": 0.0,
        "procgen-jumper": 0.0,
        "procgen-leaper": 0.0,
        "procgen-maze": 0.0,
        "procgen-miner": 0.0,
        "procgen-ninja": 0.0,
        "procgen-plunder": 0.0,
        "procgen-starpilot": 0.0,
        "crafter": 0.0,
        "miniworld-oneroom": 0.0,
        "pettingzoo-pistonball": 95.717,
        "atari-breakout": 0.0,
        "retro-airstriker": 0.0,
        "nethack-score": 0.0,
        "minihack-room": 100.0,
        "unity-food-collector": 0.0,
        "football-empty-goal": 0.0,
        "supertuxkart-lighthouse": 1.361,
        "luanti-chop-tree": 0.0,
        "supertux-first-coin": 0.0,
        "dcss-first-experience": 0.0,
        "openttd-first-road": 0.0,
        "mindustry-copper": 6.667,
        "cdda-first-weapon": 0.0,
        "warzone-first-derrick": 0.0
      },
      "errors": 0,
      "aborted_episodes": 0,
      "model": "claude-opus-5[1m]",
      "provider_metadata": {
        "transport": "authenticated-cli-per-frame",
        "requested_model": "claude-opus-5[1m]",
        "tool_events": 0,
        "observation": "png-only",
        "calls": 340,
        "prompt_version": "pixels-json-action/2",
        "resolved_models": [
          "claude-opus-5[1m]"
        ]
      },
      "hardware_details": "Apple M3 Max; 48 GiB RAM; 16 CPU cores; macOS arm64. One hosted model and one game run at a time. Each action starts a fresh authenticated CLI call. CLI startup, image transfer, network time, provider inference and action validation remain in the timer. Provider hardware is not known. Normal desktop tasks can use the same Mac."
    },
    {
      "agent": "openai-gpt-5-6-terra",
      "mode": "exhibition",
      "hardware": "Darwin arm64",
      "version": "0.5.0",
      "max_steps": 8,
      "score": 17.992,
      "ci95": null,
      "episodes": 43,
      "seeds": 1,
      "seed_ids": [
        3000
      ],
      "games": [
        "coin-run",
        "dodge-lanes",
        "doom-basic",
        "doom-defend-center",
        "doom-defend-line",
        "doom-take-cover",
        "doom-health",
        "doom-corridor",
        "doom-home",
        "doom-predict",
        "procgen-bigfish",
        "procgen-bossfight",
        "procgen-caveflyer",
        "procgen-chaser",
        "procgen-climber",
        "procgen-coinrun",
        "procgen-dodgeball",
        "procgen-fruitbot",
        "procgen-heist",
        "procgen-jumper",
        "procgen-leaper",
        "procgen-maze",
        "procgen-miner",
        "procgen-ninja",
        "procgen-plunder",
        "procgen-starpilot",
        "crafter",
        "miniworld-oneroom",
        "pettingzoo-pistonball",
        "atari-breakout",
        "retro-airstriker",
        "nethack-score",
        "minihack-room",
        "unity-food-collector",
        "football-empty-goal",
        "supertuxkart-lighthouse",
        "luanti-chop-tree",
        "supertux-first-coin",
        "dcss-first-experience",
        "openttd-first-road",
        "mindustry-copper",
        "cdda-first-weapon",
        "warzone-first-derrick"
      ],
      "task_metadata": {
        "coin-run": {},
        "dodge-lanes": {},
        "doom-basic": {},
        "doom-defend-center": {},
        "doom-defend-line": {},
        "doom-take-cover": {},
        "doom-health": {},
        "doom-corridor": {},
        "doom-home": {},
        "doom-predict": {},
        "procgen-bigfish": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            40
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bigfish",
          "task_version": "1"
        },
        "procgen-bossfight": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bossfight",
          "task_version": "1"
        },
        "procgen-caveflyer": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2,
            12
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-caveflyer",
          "task_version": "1"
        },
        "procgen-chaser": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-chaser",
          "task_version": "1"
        },
        "procgen-climber": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            12.6
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-climber",
          "task_version": "1"
        },
        "procgen-coinrun": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-coinrun",
          "task_version": "1"
        },
        "procgen-dodgeball": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            19
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-dodgeball",
          "task_version": "1"
        },
        "procgen-fruitbot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -1.5,
            32.4
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-fruitbot",
          "task_version": "1"
        },
        "procgen-heist": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-heist",
          "task_version": "1"
        },
        "procgen-jumper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-jumper",
          "task_version": "1"
        },
        "procgen-leaper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-leaper",
          "task_version": "1"
        },
        "procgen-maze": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-maze",
          "task_version": "1"
        },
        "procgen-miner": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-miner",
          "task_version": "1"
        },
        "procgen-ninja": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-ninja",
          "task_version": "1"
        },
        "procgen-plunder": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            4.5,
            30
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-plunder",
          "task_version": "1"
        },
        "procgen-starpilot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2.5,
            64
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-starpilot",
          "task_version": "1"
        },
        "crafter": {
          "engine": "crafter",
          "engine_version": "1.8.3",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            22
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "crafter",
          "task_version": "2"
        },
        "miniworld-oneroom": {
          "engine": "miniworld",
          "engine_version": "2.1.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "miniworld-oneroom",
          "task_version": "1"
        },
        "pettingzoo-pistonball": {
          "engine": "pettingzoo",
          "engine_version": "1.27.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -100,
            100
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "pettingzoo-pistonball",
          "task_version": "1"
        },
        "atari-breakout": {
          "engine": "ale-py",
          "engine_version": "0.12.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            864
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "atari-breakout",
          "task_version": "1"
        },
        "retro-airstriker": {
          "engine": "stable-retro",
          "engine_version": "1.0.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            10000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "retro-airstriker",
          "task_version": "1"
        },
        "nethack-score": {
          "engine": "nle",
          "engine_version": "1.3.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "nethack-score",
          "task_version": "1"
        },
        "minihack-room": {
          "engine": "minihack",
          "engine_version": "1.0.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "minihack-room",
          "task_version": "1"
        },
        "unity-food-collector": {
          "assets_sha256": "c28e27b2d3af4391a3e1e95b0871e01a16ef657ee062bb91e7740ac9d0c99be5",
          "camera_size": [
            84,
            84
          ],
          "capture_frame_rate": 30,
          "controlled_agent": "lowest-initial-agent-id",
          "engine": "mlagents-envs",
          "engine_version": "1.1.0",
          "numpy_version": "1.23.5",
          "observation": "rendered-rgb-only",
          "peer_policy": "no-op",
          "replay": "exact-frame-and-score",
          "reward_definition": "sum of native food rewards; green +1, red -1",
          "scene": "VisualFoodCollector",
          "score_range": [
            0,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "unity-food-collector",
          "task_version": "1"
        },
        "football-empty-goal": {
          "camera_size": [
            320,
            180
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:8774faa312fa0791f01966ef3152d23c012dda5117fd01366972a0bebfc13635",
          "control_policy": "release movement and shot before each decision",
          "engine": "gfootball",
          "engine_revision": "ba9952130898bde2879b01ff2b49d3d48d26ce81",
          "engine_version": "2.10.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native goal reward, no checkpoint shaping",
          "scenario": "academy_empty_goal_close",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "football-empty-goal",
          "task_version": "1"
        },
        "supertuxkart-lighthouse": {
          "assets_sha256": "56cd754653224efde4a40aef0400a71f03f72f18d2c35a89896db052dc096b23",
          "camera_size": [
            320,
            240
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:257166494f158e6b5b3441d4b52f99b83e8cef4ff60c6ce5b333251aa7cf4738",
          "engine": "PySuperTuxKart2",
          "engine_revision": "dd70f6823f248ae1df2ce513839a9b2c8c940c39",
          "engine_variant": "deterministic-render-v2",
          "engine_version": "0.7.4+ggbrender2",
          "neutral_preroll_steps": 50,
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "Mesa software / deterministic-render-v2",
          "renderer_threads": 1,
          "renderer_vector_bits": 128,
          "replay": "exact-frame-and-score",
          "reward_definition": "max(0, native overall distance) / native track length",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_seconds": 0.2,
          "task_id": "supertuxkart-lighthouse",
          "task_version": "3",
          "track": "lighthouse"
        },
        "luanti-chop-tree": {
          "camera_size": [
            64,
            64
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:f3f2df899b8709a904350472825f3253fd1a46a8d913322634a604e389fbb0fe",
          "engine": "craftium",
          "engine_revision": "8cffe4176e793f78d00b17fa7e46ccf333a7b5b0",
          "engine_variant": "serial-lockstep-v2",
          "engine_version": "0.0.1",
          "mesh_generation_threads": 1,
          "numpy_version": "2.5.3",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native Lua tree-node dig reward; first reward completes task",
          "scenario": "Craftium/ChopTree-v0: first tree block",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "pinned world snapshot; fixed map/Lua seed on initial reset",
          "simulation": "lockstep",
          "step_seconds": 0.02,
          "task_id": "luanti-chop-tree",
          "task_version": "1",
          "weather": "disabled via upstream enable_weather setting"
        },
        "supertux-first-coin": {
          "camera_size": [
            640,
            480
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:b3f936e8da2a38c0eab38ea1acfe56c1c47829cbe1f288775a135f0a67faadbf",
          "engine": "SuperTux",
          "engine_revision": "c1ddb4f28c54de77f07aa9965231a91c0ed06708",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.6.3+ggbench1",
          "frames_per_action": 8,
          "initialization": "skip level-introduction menu and fades; fresh player/save",
          "level": "world1/welcome_antarctica.stl",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "physics_step_seconds": 0.015,
          "renderer": "SDL software",
          "replay": "exact-frame-and-score",
          "reward_definition": "native level coin count; first coin completes task",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "supertux-first-coin",
          "task_version": "1"
        },
        "dcss-first-experience": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:942c43c70e0682b3d612e7c0b6e42eb8b35aeef42aad8bb77982bbbad4d617b8",
          "control_policy": "one native key per decision; no auto-explore or auto-fight",
          "engine": "DCSS",
          "engine_revision": "19cb0921cbcb6463fcedd1bb06636446c50b1905",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.34.0+ggbench1",
          "initialization": "advance title and seed-confirmation menus before observing",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 tiles / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native experience gain; death ends task",
          "scenario": "Dungeon:1; Minotaur Fighter with war axe; first experience",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "native uint64 seed; benchmark zero maps to uint64 maximum",
          "simulation": "lockstep",
          "task_id": "dcss-first-experience",
          "task_version": "1"
        },
        "openttd-first-road": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:227f7a13d0f6a96f2574ac0b428755101ad54def2217d629d08d059f4e2996b0",
          "cursor_step_pixels": 32,
          "engine": "OpenTTD",
          "engine_revision": "14ec60f248547d4d062a1160f0fc26d742319888",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "15.3+ggbench1",
          "initialization": "16 native ticks; fresh company and configuration",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native 32bpp software framebuffer; OpenGFX 7.1",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when native company infrastructure owns at least two road bits",
          "scenario": "new 64x64 company in 1950; first owned road",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 30,
          "task_id": "openttd-first-road",
          "task_version": "1"
        },
        "mindustry-copper": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:90468c4e5dde3dfaf18e1ec5fe337b4d32906271d3a7be899efd5b87756082a2",
          "cursor_step_pixels": 32,
          "engine": "Mindustry",
          "engine_revision": "067c720a8817c1c9fb586c03898a7d948caaed56",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "160.5+ggbench1",
          "frames_per_action": 6,
          "initialization": "120 native frames; fresh player and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "optional_effects": false,
          "physics_step_seconds": 0.016666666666666666,
          "renderer": "native SDL/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "native player-plus-core copper gain from initial total; target 15",
          "scenario": "serpulo/groundZero; native first-15-copper tutorial objective",
          "score_range": [
            0,
            15
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "mindustry-copper",
          "task_version": "1"
        },
        "cdda-first-weapon": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:eee69b35296e92b6ddc84a9ffbf737ab66ab48ae42dbbc3f1ee8e236b97c1b1f",
          "control_policy": "one native key per decision; tutorial popups remain visible",
          "engine": "Cataclysm: Dark Days Ahead",
          "engine_revision": "7b2efa5cea38e4d4d97dd0e63b28b9148623da59",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.I-1+ggbench1",
          "initialization": "native tutorial world and preset character; fresh user directory",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 software / UltimateCataclysm tiles",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when the native avatar wields the tutorial baseball bat",
          "scenario": "native tutorial; first wielded baseball bat",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "cdda-first-weapon",
          "task_version": "1"
        },
        "warzone-first-derrick": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:c2c0aee72affd4e5d5be6aa03659c533e3c968365e0f62eccf1f9fbdd30f6c22",
          "cursor_step_pixels": 32,
          "engine": "Warzone 2100",
          "engine_revision": "7d3d8e51a61ceee89e6bf672a41337e1a9cd7392",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "4.7.0+ggbench1",
          "fine_cursor_step_pixels": 8,
          "initialization": "40 native frames; fresh tutorial and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL3/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native completed structure; two unit losses end task",
          "scenario": "native TUTORIAL3; first constructed oil derrick",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 100,
          "task_id": "warzone-first-derrick",
          "task_version": "1"
        }
      },
      "decisions": 340,
      "p50_ms": 8535.438125,
      "p95_ms": 15372.738708350005,
      "max_ms": 21801.74475,
      "misses": 340,
      "latency_eligible": false,
      "latency_policy": {
        "id": "suite-p95-200-v1",
        "metric": "suite_p95_ms",
        "limit_ms": 200.0,
        "comparison": "strictly_less_than",
        "quantile_method": "linear",
        "sample_scope": "all_recorded_decisions_in_complete_suite"
      },
      "response_timeout_seconds": 120.0,
      "trust": "local-unattested",
      "official_rank": null,
      "per_game": {
        "coin-run": 100.0,
        "dodge-lanes": 100.0,
        "doom-basic": 66.19,
        "doom-defend-center": 9.524,
        "doom-defend-line": 9.524,
        "doom-take-cover": 100.0,
        "doom-health": 100.0,
        "doom-corridor": 9.091,
        "doom-home": 0.0,
        "doom-predict": 76.183,
        "procgen-bigfish": 0.0,
        "procgen-bossfight": 0.0,
        "procgen-caveflyer": 0.0,
        "procgen-chaser": 0.0,
        "procgen-climber": 0.0,
        "procgen-coinrun": 0.0,
        "procgen-dodgeball": 0.0,
        "procgen-fruitbot": 4.425,
        "procgen-heist": 0.0,
        "procgen-jumper": 0.0,
        "procgen-leaper": 0.0,
        "procgen-maze": 0.0,
        "procgen-miner": 0.0,
        "procgen-ninja": 0.0,
        "procgen-plunder": 0.0,
        "procgen-starpilot": 0.0,
        "crafter": 4.545,
        "miniworld-oneroom": 0.0,
        "pettingzoo-pistonball": 92.804,
        "atari-breakout": 0.0,
        "retro-airstriker": 0.0,
        "nethack-score": 0.0,
        "minihack-room": 100.0,
        "unity-food-collector": 0.0,
        "football-empty-goal": 0.0,
        "supertuxkart-lighthouse": 1.362,
        "luanti-chop-tree": 0.0,
        "supertux-first-coin": 0.0,
        "dcss-first-experience": 0.0,
        "openttd-first-road": 0.0,
        "mindustry-copper": 0.0,
        "cdda-first-weapon": 0.0,
        "warzone-first-derrick": 0.0
      },
      "errors": 0,
      "aborted_episodes": 0,
      "model": "gpt-5.6-terra",
      "provider_metadata": {
        "transport": "authenticated-cli-per-frame",
        "requested_model": "gpt-5.6-terra",
        "tool_events": 0,
        "observation": "png-only",
        "calls": 340,
        "prompt_version": "pixels-json-action/2"
      },
      "hardware_details": "Apple M3 Max; 48 GiB RAM; 16 CPU cores; macOS arm64. One hosted model and one game run at a time. Each action starts a fresh authenticated CLI call. CLI startup, image transfer, network time, provider inference and action validation remain in the timer. Provider hardware is not known. Normal desktop tasks can use the same Mac."
    },
    {
      "agent": "claude-claude-sonnet-5",
      "mode": "exhibition",
      "hardware": "Darwin arm64",
      "version": "0.5.0",
      "max_steps": 8,
      "score": 17.757,
      "ci95": null,
      "episodes": 43,
      "seeds": 1,
      "seed_ids": [
        3000
      ],
      "games": [
        "coin-run",
        "dodge-lanes",
        "doom-basic",
        "doom-defend-center",
        "doom-defend-line",
        "doom-take-cover",
        "doom-health",
        "doom-corridor",
        "doom-home",
        "doom-predict",
        "procgen-bigfish",
        "procgen-bossfight",
        "procgen-caveflyer",
        "procgen-chaser",
        "procgen-climber",
        "procgen-coinrun",
        "procgen-dodgeball",
        "procgen-fruitbot",
        "procgen-heist",
        "procgen-jumper",
        "procgen-leaper",
        "procgen-maze",
        "procgen-miner",
        "procgen-ninja",
        "procgen-plunder",
        "procgen-starpilot",
        "crafter",
        "miniworld-oneroom",
        "pettingzoo-pistonball",
        "atari-breakout",
        "retro-airstriker",
        "nethack-score",
        "minihack-room",
        "unity-food-collector",
        "football-empty-goal",
        "supertuxkart-lighthouse",
        "luanti-chop-tree",
        "supertux-first-coin",
        "dcss-first-experience",
        "openttd-first-road",
        "mindustry-copper",
        "cdda-first-weapon",
        "warzone-first-derrick"
      ],
      "task_metadata": {
        "coin-run": {},
        "dodge-lanes": {},
        "doom-basic": {},
        "doom-defend-center": {},
        "doom-defend-line": {},
        "doom-take-cover": {},
        "doom-health": {},
        "doom-corridor": {},
        "doom-home": {},
        "doom-predict": {},
        "procgen-bigfish": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            40
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bigfish",
          "task_version": "1"
        },
        "procgen-bossfight": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bossfight",
          "task_version": "1"
        },
        "procgen-caveflyer": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2,
            12
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-caveflyer",
          "task_version": "1"
        },
        "procgen-chaser": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-chaser",
          "task_version": "1"
        },
        "procgen-climber": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            12.6
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-climber",
          "task_version": "1"
        },
        "procgen-coinrun": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-coinrun",
          "task_version": "1"
        },
        "procgen-dodgeball": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            19
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-dodgeball",
          "task_version": "1"
        },
        "procgen-fruitbot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -1.5,
            32.4
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-fruitbot",
          "task_version": "1"
        },
        "procgen-heist": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-heist",
          "task_version": "1"
        },
        "procgen-jumper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-jumper",
          "task_version": "1"
        },
        "procgen-leaper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-leaper",
          "task_version": "1"
        },
        "procgen-maze": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-maze",
          "task_version": "1"
        },
        "procgen-miner": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-miner",
          "task_version": "1"
        },
        "procgen-ninja": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-ninja",
          "task_version": "1"
        },
        "procgen-plunder": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            4.5,
            30
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-plunder",
          "task_version": "1"
        },
        "procgen-starpilot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2.5,
            64
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-starpilot",
          "task_version": "1"
        },
        "crafter": {
          "engine": "crafter",
          "engine_version": "1.8.3",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            22
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "crafter",
          "task_version": "2"
        },
        "miniworld-oneroom": {
          "engine": "miniworld",
          "engine_version": "2.1.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "miniworld-oneroom",
          "task_version": "1"
        },
        "pettingzoo-pistonball": {
          "engine": "pettingzoo",
          "engine_version": "1.27.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -100,
            100
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "pettingzoo-pistonball",
          "task_version": "1"
        },
        "atari-breakout": {
          "engine": "ale-py",
          "engine_version": "0.12.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            864
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "atari-breakout",
          "task_version": "1"
        },
        "retro-airstriker": {
          "engine": "stable-retro",
          "engine_version": "1.0.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            10000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "retro-airstriker",
          "task_version": "1"
        },
        "nethack-score": {
          "engine": "nle",
          "engine_version": "1.3.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "nethack-score",
          "task_version": "1"
        },
        "minihack-room": {
          "engine": "minihack",
          "engine_version": "1.0.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "minihack-room",
          "task_version": "1"
        },
        "unity-food-collector": {
          "assets_sha256": "c28e27b2d3af4391a3e1e95b0871e01a16ef657ee062bb91e7740ac9d0c99be5",
          "camera_size": [
            84,
            84
          ],
          "capture_frame_rate": 30,
          "controlled_agent": "lowest-initial-agent-id",
          "engine": "mlagents-envs",
          "engine_version": "1.1.0",
          "numpy_version": "1.23.5",
          "observation": "rendered-rgb-only",
          "peer_policy": "no-op",
          "replay": "exact-frame-and-score",
          "reward_definition": "sum of native food rewards; green +1, red -1",
          "scene": "VisualFoodCollector",
          "score_range": [
            0,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "unity-food-collector",
          "task_version": "1"
        },
        "football-empty-goal": {
          "camera_size": [
            320,
            180
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:8774faa312fa0791f01966ef3152d23c012dda5117fd01366972a0bebfc13635",
          "control_policy": "release movement and shot before each decision",
          "engine": "gfootball",
          "engine_revision": "ba9952130898bde2879b01ff2b49d3d48d26ce81",
          "engine_version": "2.10.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native goal reward, no checkpoint shaping",
          "scenario": "academy_empty_goal_close",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "football-empty-goal",
          "task_version": "1"
        },
        "supertuxkart-lighthouse": {
          "assets_sha256": "56cd754653224efde4a40aef0400a71f03f72f18d2c35a89896db052dc096b23",
          "camera_size": [
            320,
            240
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:257166494f158e6b5b3441d4b52f99b83e8cef4ff60c6ce5b333251aa7cf4738",
          "engine": "PySuperTuxKart2",
          "engine_revision": "dd70f6823f248ae1df2ce513839a9b2c8c940c39",
          "engine_variant": "deterministic-render-v2",
          "engine_version": "0.7.4+ggbrender2",
          "neutral_preroll_steps": 50,
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "Mesa software / deterministic-render-v2",
          "renderer_threads": 1,
          "renderer_vector_bits": 128,
          "replay": "exact-frame-and-score",
          "reward_definition": "max(0, native overall distance) / native track length",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_seconds": 0.2,
          "task_id": "supertuxkart-lighthouse",
          "task_version": "3",
          "track": "lighthouse"
        },
        "luanti-chop-tree": {
          "camera_size": [
            64,
            64
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:f3f2df899b8709a904350472825f3253fd1a46a8d913322634a604e389fbb0fe",
          "engine": "craftium",
          "engine_revision": "8cffe4176e793f78d00b17fa7e46ccf333a7b5b0",
          "engine_variant": "serial-lockstep-v2",
          "engine_version": "0.0.1",
          "mesh_generation_threads": 1,
          "numpy_version": "2.5.3",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native Lua tree-node dig reward; first reward completes task",
          "scenario": "Craftium/ChopTree-v0: first tree block",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "pinned world snapshot; fixed map/Lua seed on initial reset",
          "simulation": "lockstep",
          "step_seconds": 0.02,
          "task_id": "luanti-chop-tree",
          "task_version": "1",
          "weather": "disabled via upstream enable_weather setting"
        },
        "supertux-first-coin": {
          "camera_size": [
            640,
            480
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:b3f936e8da2a38c0eab38ea1acfe56c1c47829cbe1f288775a135f0a67faadbf",
          "engine": "SuperTux",
          "engine_revision": "c1ddb4f28c54de77f07aa9965231a91c0ed06708",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.6.3+ggbench1",
          "frames_per_action": 8,
          "initialization": "skip level-introduction menu and fades; fresh player/save",
          "level": "world1/welcome_antarctica.stl",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "physics_step_seconds": 0.015,
          "renderer": "SDL software",
          "replay": "exact-frame-and-score",
          "reward_definition": "native level coin count; first coin completes task",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "supertux-first-coin",
          "task_version": "1"
        },
        "dcss-first-experience": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:942c43c70e0682b3d612e7c0b6e42eb8b35aeef42aad8bb77982bbbad4d617b8",
          "control_policy": "one native key per decision; no auto-explore or auto-fight",
          "engine": "DCSS",
          "engine_revision": "19cb0921cbcb6463fcedd1bb06636446c50b1905",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.34.0+ggbench1",
          "initialization": "advance title and seed-confirmation menus before observing",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 tiles / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native experience gain; death ends task",
          "scenario": "Dungeon:1; Minotaur Fighter with war axe; first experience",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "native uint64 seed; benchmark zero maps to uint64 maximum",
          "simulation": "lockstep",
          "task_id": "dcss-first-experience",
          "task_version": "1"
        },
        "openttd-first-road": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:227f7a13d0f6a96f2574ac0b428755101ad54def2217d629d08d059f4e2996b0",
          "cursor_step_pixels": 32,
          "engine": "OpenTTD",
          "engine_revision": "14ec60f248547d4d062a1160f0fc26d742319888",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "15.3+ggbench1",
          "initialization": "16 native ticks; fresh company and configuration",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native 32bpp software framebuffer; OpenGFX 7.1",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when native company infrastructure owns at least two road bits",
          "scenario": "new 64x64 company in 1950; first owned road",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 30,
          "task_id": "openttd-first-road",
          "task_version": "1"
        },
        "mindustry-copper": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:90468c4e5dde3dfaf18e1ec5fe337b4d32906271d3a7be899efd5b87756082a2",
          "cursor_step_pixels": 32,
          "engine": "Mindustry",
          "engine_revision": "067c720a8817c1c9fb586c03898a7d948caaed56",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "160.5+ggbench1",
          "frames_per_action": 6,
          "initialization": "120 native frames; fresh player and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "optional_effects": false,
          "physics_step_seconds": 0.016666666666666666,
          "renderer": "native SDL/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "native player-plus-core copper gain from initial total; target 15",
          "scenario": "serpulo/groundZero; native first-15-copper tutorial objective",
          "score_range": [
            0,
            15
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "mindustry-copper",
          "task_version": "1"
        },
        "cdda-first-weapon": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:eee69b35296e92b6ddc84a9ffbf737ab66ab48ae42dbbc3f1ee8e236b97c1b1f",
          "control_policy": "one native key per decision; tutorial popups remain visible",
          "engine": "Cataclysm: Dark Days Ahead",
          "engine_revision": "7b2efa5cea38e4d4d97dd0e63b28b9148623da59",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.I-1+ggbench1",
          "initialization": "native tutorial world and preset character; fresh user directory",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 software / UltimateCataclysm tiles",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when the native avatar wields the tutorial baseball bat",
          "scenario": "native tutorial; first wielded baseball bat",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "cdda-first-weapon",
          "task_version": "1"
        },
        "warzone-first-derrick": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:c2c0aee72affd4e5d5be6aa03659c533e3c968365e0f62eccf1f9fbdd30f6c22",
          "cursor_step_pixels": 32,
          "engine": "Warzone 2100",
          "engine_revision": "7d3d8e51a61ceee89e6bf672a41337e1a9cd7392",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "4.7.0+ggbench1",
          "fine_cursor_step_pixels": 8,
          "initialization": "40 native frames; fresh tutorial and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL3/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native completed structure; two unit losses end task",
          "scenario": "native TUTORIAL3; first constructed oil derrick",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 100,
          "task_id": "warzone-first-derrick",
          "task_version": "1"
        }
      },
      "decisions": 340,
      "p50_ms": 1998.6229375,
      "p95_ms": 2761.4562360500004,
      "max_ms": 8639.477125,
      "misses": 340,
      "latency_eligible": false,
      "latency_policy": {
        "id": "suite-p95-200-v1",
        "metric": "suite_p95_ms",
        "limit_ms": 200.0,
        "comparison": "strictly_less_than",
        "quantile_method": "linear",
        "sample_scope": "all_recorded_decisions_in_complete_suite"
      },
      "response_timeout_seconds": 120.0,
      "trust": "local-unattested",
      "official_rank": null,
      "per_game": {
        "coin-run": 100.0,
        "dodge-lanes": 100.0,
        "doom-basic": 66.19,
        "doom-defend-center": 4.762,
        "doom-defend-line": 9.524,
        "doom-take-cover": 100.0,
        "doom-health": 100.0,
        "doom-corridor": 9.091,
        "doom-home": 0.0,
        "doom-predict": 76.183,
        "procgen-bigfish": 0.0,
        "procgen-bossfight": 0.0,
        "procgen-caveflyer": 0.0,
        "procgen-chaser": 0.0,
        "procgen-climber": 0.0,
        "procgen-coinrun": 0.0,
        "procgen-dodgeball": 0.0,
        "procgen-fruitbot": 4.425,
        "procgen-heist": 0.0,
        "procgen-jumper": 0.0,
        "procgen-leaper": 0.0,
        "procgen-maze": 0.0,
        "procgen-miner": 0.0,
        "procgen-ninja": 0.0,
        "procgen-plunder": 0.0,
        "procgen-starpilot": 0.0,
        "crafter": 4.545,
        "miniworld-oneroom": 0.0,
        "pettingzoo-pistonball": 87.464,
        "atari-breakout": 0.0,
        "retro-airstriker": 0.0,
        "nethack-score": 0.0,
        "minihack-room": 100.0,
        "unity-food-collector": 0.0,
        "football-empty-goal": 0.0,
        "supertuxkart-lighthouse": 1.362,
        "luanti-chop-tree": 0.0,
        "supertux-first-coin": 0.0,
        "dcss-first-experience": 0.0,
        "openttd-first-road": 0.0,
        "mindustry-copper": 0.0,
        "cdda-first-weapon": 0.0,
        "warzone-first-derrick": 0.0
      },
      "errors": 0,
      "aborted_episodes": 0,
      "model": "claude-sonnet-5",
      "provider_metadata": {
        "transport": "authenticated-cli-per-frame",
        "requested_model": "claude-sonnet-5",
        "tool_events": 0,
        "observation": "png-only",
        "calls": 340,
        "prompt_version": "pixels-json-action/2",
        "resolved_models": [
          "claude-sonnet-5"
        ]
      },
      "hardware_details": "Apple M3 Max; 48 GiB RAM; 16 CPU cores; macOS arm64. One hosted model and one game run at a time. Each action starts a fresh authenticated CLI call. CLI startup, image transfer, network time, provider inference and action validation remain in the timer. Provider hardware is not known. Normal desktop tasks can use the same Mac."
    },
    {
      "agent": "openai-gpt-5-5",
      "mode": "exhibition",
      "hardware": "Darwin arm64",
      "version": "0.5.0",
      "max_steps": 8,
      "score": 17.344,
      "ci95": null,
      "episodes": 43,
      "seeds": 1,
      "seed_ids": [
        3000
      ],
      "games": [
        "coin-run",
        "dodge-lanes",
        "doom-basic",
        "doom-defend-center",
        "doom-defend-line",
        "doom-take-cover",
        "doom-health",
        "doom-corridor",
        "doom-home",
        "doom-predict",
        "procgen-bigfish",
        "procgen-bossfight",
        "procgen-caveflyer",
        "procgen-chaser",
        "procgen-climber",
        "procgen-coinrun",
        "procgen-dodgeball",
        "procgen-fruitbot",
        "procgen-heist",
        "procgen-jumper",
        "procgen-leaper",
        "procgen-maze",
        "procgen-miner",
        "procgen-ninja",
        "procgen-plunder",
        "procgen-starpilot",
        "crafter",
        "miniworld-oneroom",
        "pettingzoo-pistonball",
        "atari-breakout",
        "retro-airstriker",
        "nethack-score",
        "minihack-room",
        "unity-food-collector",
        "football-empty-goal",
        "supertuxkart-lighthouse",
        "luanti-chop-tree",
        "supertux-first-coin",
        "dcss-first-experience",
        "openttd-first-road",
        "mindustry-copper",
        "cdda-first-weapon",
        "warzone-first-derrick"
      ],
      "task_metadata": {
        "coin-run": {},
        "dodge-lanes": {},
        "doom-basic": {},
        "doom-defend-center": {},
        "doom-defend-line": {},
        "doom-take-cover": {},
        "doom-health": {},
        "doom-corridor": {},
        "doom-home": {},
        "doom-predict": {},
        "procgen-bigfish": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            40
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bigfish",
          "task_version": "1"
        },
        "procgen-bossfight": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bossfight",
          "task_version": "1"
        },
        "procgen-caveflyer": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2,
            12
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-caveflyer",
          "task_version": "1"
        },
        "procgen-chaser": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-chaser",
          "task_version": "1"
        },
        "procgen-climber": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            12.6
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-climber",
          "task_version": "1"
        },
        "procgen-coinrun": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-coinrun",
          "task_version": "1"
        },
        "procgen-dodgeball": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            19
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-dodgeball",
          "task_version": "1"
        },
        "procgen-fruitbot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -1.5,
            32.4
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-fruitbot",
          "task_version": "1"
        },
        "procgen-heist": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-heist",
          "task_version": "1"
        },
        "procgen-jumper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-jumper",
          "task_version": "1"
        },
        "procgen-leaper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-leaper",
          "task_version": "1"
        },
        "procgen-maze": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-maze",
          "task_version": "1"
        },
        "procgen-miner": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-miner",
          "task_version": "1"
        },
        "procgen-ninja": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-ninja",
          "task_version": "1"
        },
        "procgen-plunder": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            4.5,
            30
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-plunder",
          "task_version": "1"
        },
        "procgen-starpilot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2.5,
            64
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-starpilot",
          "task_version": "1"
        },
        "crafter": {
          "engine": "crafter",
          "engine_version": "1.8.3",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            22
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "crafter",
          "task_version": "2"
        },
        "miniworld-oneroom": {
          "engine": "miniworld",
          "engine_version": "2.1.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "miniworld-oneroom",
          "task_version": "1"
        },
        "pettingzoo-pistonball": {
          "engine": "pettingzoo",
          "engine_version": "1.27.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -100,
            100
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "pettingzoo-pistonball",
          "task_version": "1"
        },
        "atari-breakout": {
          "engine": "ale-py",
          "engine_version": "0.12.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            864
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "atari-breakout",
          "task_version": "1"
        },
        "retro-airstriker": {
          "engine": "stable-retro",
          "engine_version": "1.0.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            10000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "retro-airstriker",
          "task_version": "1"
        },
        "nethack-score": {
          "engine": "nle",
          "engine_version": "1.3.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "nethack-score",
          "task_version": "1"
        },
        "minihack-room": {
          "engine": "minihack",
          "engine_version": "1.0.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "minihack-room",
          "task_version": "1"
        },
        "unity-food-collector": {
          "assets_sha256": "c28e27b2d3af4391a3e1e95b0871e01a16ef657ee062bb91e7740ac9d0c99be5",
          "camera_size": [
            84,
            84
          ],
          "capture_frame_rate": 30,
          "controlled_agent": "lowest-initial-agent-id",
          "engine": "mlagents-envs",
          "engine_version": "1.1.0",
          "numpy_version": "1.23.5",
          "observation": "rendered-rgb-only",
          "peer_policy": "no-op",
          "replay": "exact-frame-and-score",
          "reward_definition": "sum of native food rewards; green +1, red -1",
          "scene": "VisualFoodCollector",
          "score_range": [
            0,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "unity-food-collector",
          "task_version": "1"
        },
        "football-empty-goal": {
          "camera_size": [
            320,
            180
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:8774faa312fa0791f01966ef3152d23c012dda5117fd01366972a0bebfc13635",
          "control_policy": "release movement and shot before each decision",
          "engine": "gfootball",
          "engine_revision": "ba9952130898bde2879b01ff2b49d3d48d26ce81",
          "engine_version": "2.10.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native goal reward, no checkpoint shaping",
          "scenario": "academy_empty_goal_close",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "football-empty-goal",
          "task_version": "1"
        },
        "supertuxkart-lighthouse": {
          "assets_sha256": "56cd754653224efde4a40aef0400a71f03f72f18d2c35a89896db052dc096b23",
          "camera_size": [
            320,
            240
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:257166494f158e6b5b3441d4b52f99b83e8cef4ff60c6ce5b333251aa7cf4738",
          "engine": "PySuperTuxKart2",
          "engine_revision": "dd70f6823f248ae1df2ce513839a9b2c8c940c39",
          "engine_variant": "deterministic-render-v2",
          "engine_version": "0.7.4+ggbrender2",
          "neutral_preroll_steps": 50,
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "Mesa software / deterministic-render-v2",
          "renderer_threads": 1,
          "renderer_vector_bits": 128,
          "replay": "exact-frame-and-score",
          "reward_definition": "max(0, native overall distance) / native track length",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_seconds": 0.2,
          "task_id": "supertuxkart-lighthouse",
          "task_version": "3",
          "track": "lighthouse"
        },
        "luanti-chop-tree": {
          "camera_size": [
            64,
            64
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:f3f2df899b8709a904350472825f3253fd1a46a8d913322634a604e389fbb0fe",
          "engine": "craftium",
          "engine_revision": "8cffe4176e793f78d00b17fa7e46ccf333a7b5b0",
          "engine_variant": "serial-lockstep-v2",
          "engine_version": "0.0.1",
          "mesh_generation_threads": 1,
          "numpy_version": "2.5.3",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native Lua tree-node dig reward; first reward completes task",
          "scenario": "Craftium/ChopTree-v0: first tree block",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "pinned world snapshot; fixed map/Lua seed on initial reset",
          "simulation": "lockstep",
          "step_seconds": 0.02,
          "task_id": "luanti-chop-tree",
          "task_version": "1",
          "weather": "disabled via upstream enable_weather setting"
        },
        "supertux-first-coin": {
          "camera_size": [
            640,
            480
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:b3f936e8da2a38c0eab38ea1acfe56c1c47829cbe1f288775a135f0a67faadbf",
          "engine": "SuperTux",
          "engine_revision": "c1ddb4f28c54de77f07aa9965231a91c0ed06708",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.6.3+ggbench1",
          "frames_per_action": 8,
          "initialization": "skip level-introduction menu and fades; fresh player/save",
          "level": "world1/welcome_antarctica.stl",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "physics_step_seconds": 0.015,
          "renderer": "SDL software",
          "replay": "exact-frame-and-score",
          "reward_definition": "native level coin count; first coin completes task",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "supertux-first-coin",
          "task_version": "1"
        },
        "dcss-first-experience": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:942c43c70e0682b3d612e7c0b6e42eb8b35aeef42aad8bb77982bbbad4d617b8",
          "control_policy": "one native key per decision; no auto-explore or auto-fight",
          "engine": "DCSS",
          "engine_revision": "19cb0921cbcb6463fcedd1bb06636446c50b1905",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.34.0+ggbench1",
          "initialization": "advance title and seed-confirmation menus before observing",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 tiles / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native experience gain; death ends task",
          "scenario": "Dungeon:1; Minotaur Fighter with war axe; first experience",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "native uint64 seed; benchmark zero maps to uint64 maximum",
          "simulation": "lockstep",
          "task_id": "dcss-first-experience",
          "task_version": "1"
        },
        "openttd-first-road": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:227f7a13d0f6a96f2574ac0b428755101ad54def2217d629d08d059f4e2996b0",
          "cursor_step_pixels": 32,
          "engine": "OpenTTD",
          "engine_revision": "14ec60f248547d4d062a1160f0fc26d742319888",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "15.3+ggbench1",
          "initialization": "16 native ticks; fresh company and configuration",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native 32bpp software framebuffer; OpenGFX 7.1",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when native company infrastructure owns at least two road bits",
          "scenario": "new 64x64 company in 1950; first owned road",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 30,
          "task_id": "openttd-first-road",
          "task_version": "1"
        },
        "mindustry-copper": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:90468c4e5dde3dfaf18e1ec5fe337b4d32906271d3a7be899efd5b87756082a2",
          "cursor_step_pixels": 32,
          "engine": "Mindustry",
          "engine_revision": "067c720a8817c1c9fb586c03898a7d948caaed56",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "160.5+ggbench1",
          "frames_per_action": 6,
          "initialization": "120 native frames; fresh player and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "optional_effects": false,
          "physics_step_seconds": 0.016666666666666666,
          "renderer": "native SDL/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "native player-plus-core copper gain from initial total; target 15",
          "scenario": "serpulo/groundZero; native first-15-copper tutorial objective",
          "score_range": [
            0,
            15
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "mindustry-copper",
          "task_version": "1"
        },
        "cdda-first-weapon": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:eee69b35296e92b6ddc84a9ffbf737ab66ab48ae42dbbc3f1ee8e236b97c1b1f",
          "control_policy": "one native key per decision; tutorial popups remain visible",
          "engine": "Cataclysm: Dark Days Ahead",
          "engine_revision": "7b2efa5cea38e4d4d97dd0e63b28b9148623da59",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.I-1+ggbench1",
          "initialization": "native tutorial world and preset character; fresh user directory",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 software / UltimateCataclysm tiles",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when the native avatar wields the tutorial baseball bat",
          "scenario": "native tutorial; first wielded baseball bat",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "cdda-first-weapon",
          "task_version": "1"
        },
        "warzone-first-derrick": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:c2c0aee72affd4e5d5be6aa03659c533e3c968365e0f62eccf1f9fbdd30f6c22",
          "cursor_step_pixels": 32,
          "engine": "Warzone 2100",
          "engine_revision": "7d3d8e51a61ceee89e6bf672a41337e1a9cd7392",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "4.7.0+ggbench1",
          "fine_cursor_step_pixels": 8,
          "initialization": "40 native frames; fresh tutorial and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL3/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native completed structure; two unit losses end task",
          "scenario": "native TUTORIAL3; first constructed oil derrick",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 100,
          "task_id": "warzone-first-derrick",
          "task_version": "1"
        }
      },
      "decisions": 340,
      "p50_ms": 6897.5903545,
      "p95_ms": 12131.91553605,
      "max_ms": 16019.811625,
      "misses": 340,
      "latency_eligible": false,
      "latency_policy": {
        "id": "suite-p95-200-v1",
        "metric": "suite_p95_ms",
        "limit_ms": 200.0,
        "comparison": "strictly_less_than",
        "quantile_method": "linear",
        "sample_scope": "all_recorded_decisions_in_complete_suite"
      },
      "response_timeout_seconds": 120.0,
      "trust": "local-unattested",
      "official_rank": null,
      "per_game": {
        "coin-run": 100.0,
        "dodge-lanes": 100.0,
        "doom-basic": 66.19,
        "doom-defend-center": 9.524,
        "doom-defend-line": 9.524,
        "doom-take-cover": 100.0,
        "doom-health": 100.0,
        "doom-corridor": 9.091,
        "doom-home": 0.0,
        "doom-predict": 76.183,
        "procgen-bigfish": 0.0,
        "procgen-bossfight": 0.0,
        "procgen-caveflyer": 0.0,
        "procgen-chaser": 0.0,
        "procgen-climber": 0.0,
        "procgen-coinrun": 0.0,
        "procgen-dodgeball": 0.0,
        "procgen-fruitbot": 4.425,
        "procgen-heist": 0.0,
        "procgen-jumper": 0.0,
        "procgen-leaper": 0.0,
        "procgen-maze": 0.0,
        "procgen-miner": 0.0,
        "procgen-ninja": 0.0,
        "procgen-plunder": 0.0,
        "procgen-starpilot": 0.0,
        "crafter": 0.0,
        "miniworld-oneroom": 0.0,
        "pettingzoo-pistonball": 69.503,
        "atari-breakout": 0.0,
        "retro-airstriker": 0.0,
        "nethack-score": 0.0,
        "minihack-room": 100.0,
        "unity-food-collector": 0.0,
        "football-empty-goal": 0.0,
        "supertuxkart-lighthouse": 1.34,
        "luanti-chop-tree": 0.0,
        "supertux-first-coin": 0.0,
        "dcss-first-experience": 0.0,
        "openttd-first-road": 0.0,
        "mindustry-copper": 0.0,
        "cdda-first-weapon": 0.0,
        "warzone-first-derrick": 0.0
      },
      "errors": 0,
      "aborted_episodes": 0,
      "model": "gpt-5.5",
      "provider_metadata": {
        "transport": "authenticated-cli-per-frame",
        "requested_model": "gpt-5.5",
        "tool_events": 0,
        "observation": "png-only",
        "calls": 340,
        "prompt_version": "pixels-json-action/2"
      },
      "hardware_details": "Apple M3 Max; 48 GiB RAM; 16 CPU cores; macOS arm64. One hosted model and one game run at a time. Each action starts a fresh authenticated CLI call. CLI startup, image transfer, network time, provider inference and action validation remain in the timer. Provider hardware is not known. Normal desktop tasks can use the same Mac."
    },
    {
      "agent": "openai-gpt-6-sol",
      "mode": "exhibition",
      "hardware": "Darwin arm64",
      "version": "0.5.0",
      "max_steps": 8,
      "score": 15.703,
      "ci95": null,
      "episodes": 43,
      "seeds": 1,
      "seed_ids": [
        3000
      ],
      "games": [
        "coin-run",
        "dodge-lanes",
        "doom-basic",
        "doom-defend-center",
        "doom-defend-line",
        "doom-take-cover",
        "doom-health",
        "doom-corridor",
        "doom-home",
        "doom-predict",
        "procgen-bigfish",
        "procgen-bossfight",
        "procgen-caveflyer",
        "procgen-chaser",
        "procgen-climber",
        "procgen-coinrun",
        "procgen-dodgeball",
        "procgen-fruitbot",
        "procgen-heist",
        "procgen-jumper",
        "procgen-leaper",
        "procgen-maze",
        "procgen-miner",
        "procgen-ninja",
        "procgen-plunder",
        "procgen-starpilot",
        "crafter",
        "miniworld-oneroom",
        "pettingzoo-pistonball",
        "atari-breakout",
        "retro-airstriker",
        "nethack-score",
        "minihack-room",
        "unity-food-collector",
        "football-empty-goal",
        "supertuxkart-lighthouse",
        "luanti-chop-tree",
        "supertux-first-coin",
        "dcss-first-experience",
        "openttd-first-road",
        "mindustry-copper",
        "cdda-first-weapon",
        "warzone-first-derrick"
      ],
      "task_metadata": {
        "coin-run": {},
        "dodge-lanes": {},
        "doom-basic": {},
        "doom-defend-center": {},
        "doom-defend-line": {},
        "doom-take-cover": {},
        "doom-health": {},
        "doom-corridor": {},
        "doom-home": {},
        "doom-predict": {},
        "procgen-bigfish": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            40
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bigfish",
          "task_version": "1"
        },
        "procgen-bossfight": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bossfight",
          "task_version": "1"
        },
        "procgen-caveflyer": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2,
            12
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-caveflyer",
          "task_version": "1"
        },
        "procgen-chaser": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-chaser",
          "task_version": "1"
        },
        "procgen-climber": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            12.6
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-climber",
          "task_version": "1"
        },
        "procgen-coinrun": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-coinrun",
          "task_version": "1"
        },
        "procgen-dodgeball": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            19
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-dodgeball",
          "task_version": "1"
        },
        "procgen-fruitbot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -1.5,
            32.4
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-fruitbot",
          "task_version": "1"
        },
        "procgen-heist": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-heist",
          "task_version": "1"
        },
        "procgen-jumper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-jumper",
          "task_version": "1"
        },
        "procgen-leaper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-leaper",
          "task_version": "1"
        },
        "procgen-maze": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-maze",
          "task_version": "1"
        },
        "procgen-miner": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-miner",
          "task_version": "1"
        },
        "procgen-ninja": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-ninja",
          "task_version": "1"
        },
        "procgen-plunder": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            4.5,
            30
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-plunder",
          "task_version": "1"
        },
        "procgen-starpilot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2.5,
            64
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-starpilot",
          "task_version": "1"
        },
        "crafter": {
          "engine": "crafter",
          "engine_version": "1.8.3",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            22
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "crafter",
          "task_version": "2"
        },
        "miniworld-oneroom": {
          "engine": "miniworld",
          "engine_version": "2.1.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "miniworld-oneroom",
          "task_version": "1"
        },
        "pettingzoo-pistonball": {
          "engine": "pettingzoo",
          "engine_version": "1.27.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -100,
            100
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "pettingzoo-pistonball",
          "task_version": "1"
        },
        "atari-breakout": {
          "engine": "ale-py",
          "engine_version": "0.12.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            864
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "atari-breakout",
          "task_version": "1"
        },
        "retro-airstriker": {
          "engine": "stable-retro",
          "engine_version": "1.0.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            10000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "retro-airstriker",
          "task_version": "1"
        },
        "nethack-score": {
          "engine": "nle",
          "engine_version": "1.3.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "nethack-score",
          "task_version": "1"
        },
        "minihack-room": {
          "engine": "minihack",
          "engine_version": "1.0.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "minihack-room",
          "task_version": "1"
        },
        "unity-food-collector": {
          "assets_sha256": "c28e27b2d3af4391a3e1e95b0871e01a16ef657ee062bb91e7740ac9d0c99be5",
          "camera_size": [
            84,
            84
          ],
          "capture_frame_rate": 30,
          "controlled_agent": "lowest-initial-agent-id",
          "engine": "mlagents-envs",
          "engine_version": "1.1.0",
          "numpy_version": "1.23.5",
          "observation": "rendered-rgb-only",
          "peer_policy": "no-op",
          "replay": "exact-frame-and-score",
          "reward_definition": "sum of native food rewards; green +1, red -1",
          "scene": "VisualFoodCollector",
          "score_range": [
            0,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "unity-food-collector",
          "task_version": "1"
        },
        "football-empty-goal": {
          "camera_size": [
            320,
            180
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:8774faa312fa0791f01966ef3152d23c012dda5117fd01366972a0bebfc13635",
          "control_policy": "release movement and shot before each decision",
          "engine": "gfootball",
          "engine_revision": "ba9952130898bde2879b01ff2b49d3d48d26ce81",
          "engine_version": "2.10.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native goal reward, no checkpoint shaping",
          "scenario": "academy_empty_goal_close",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "football-empty-goal",
          "task_version": "1"
        },
        "supertuxkart-lighthouse": {
          "assets_sha256": "56cd754653224efde4a40aef0400a71f03f72f18d2c35a89896db052dc096b23",
          "camera_size": [
            320,
            240
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:257166494f158e6b5b3441d4b52f99b83e8cef4ff60c6ce5b333251aa7cf4738",
          "engine": "PySuperTuxKart2",
          "engine_revision": "dd70f6823f248ae1df2ce513839a9b2c8c940c39",
          "engine_variant": "deterministic-render-v2",
          "engine_version": "0.7.4+ggbrender2",
          "neutral_preroll_steps": 50,
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "Mesa software / deterministic-render-v2",
          "renderer_threads": 1,
          "renderer_vector_bits": 128,
          "replay": "exact-frame-and-score",
          "reward_definition": "max(0, native overall distance) / native track length",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_seconds": 0.2,
          "task_id": "supertuxkart-lighthouse",
          "task_version": "3",
          "track": "lighthouse"
        },
        "luanti-chop-tree": {
          "camera_size": [
            64,
            64
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:f3f2df899b8709a904350472825f3253fd1a46a8d913322634a604e389fbb0fe",
          "engine": "craftium",
          "engine_revision": "8cffe4176e793f78d00b17fa7e46ccf333a7b5b0",
          "engine_variant": "serial-lockstep-v2",
          "engine_version": "0.0.1",
          "mesh_generation_threads": 1,
          "numpy_version": "2.5.3",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native Lua tree-node dig reward; first reward completes task",
          "scenario": "Craftium/ChopTree-v0: first tree block",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "pinned world snapshot; fixed map/Lua seed on initial reset",
          "simulation": "lockstep",
          "step_seconds": 0.02,
          "task_id": "luanti-chop-tree",
          "task_version": "1",
          "weather": "disabled via upstream enable_weather setting"
        },
        "supertux-first-coin": {
          "camera_size": [
            640,
            480
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:b3f936e8da2a38c0eab38ea1acfe56c1c47829cbe1f288775a135f0a67faadbf",
          "engine": "SuperTux",
          "engine_revision": "c1ddb4f28c54de77f07aa9965231a91c0ed06708",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.6.3+ggbench1",
          "frames_per_action": 8,
          "initialization": "skip level-introduction menu and fades; fresh player/save",
          "level": "world1/welcome_antarctica.stl",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "physics_step_seconds": 0.015,
          "renderer": "SDL software",
          "replay": "exact-frame-and-score",
          "reward_definition": "native level coin count; first coin completes task",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "supertux-first-coin",
          "task_version": "1"
        },
        "dcss-first-experience": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:942c43c70e0682b3d612e7c0b6e42eb8b35aeef42aad8bb77982bbbad4d617b8",
          "control_policy": "one native key per decision; no auto-explore or auto-fight",
          "engine": "DCSS",
          "engine_revision": "19cb0921cbcb6463fcedd1bb06636446c50b1905",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.34.0+ggbench1",
          "initialization": "advance title and seed-confirmation menus before observing",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 tiles / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native experience gain; death ends task",
          "scenario": "Dungeon:1; Minotaur Fighter with war axe; first experience",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "native uint64 seed; benchmark zero maps to uint64 maximum",
          "simulation": "lockstep",
          "task_id": "dcss-first-experience",
          "task_version": "1"
        },
        "openttd-first-road": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:227f7a13d0f6a96f2574ac0b428755101ad54def2217d629d08d059f4e2996b0",
          "cursor_step_pixels": 32,
          "engine": "OpenTTD",
          "engine_revision": "14ec60f248547d4d062a1160f0fc26d742319888",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "15.3+ggbench1",
          "initialization": "16 native ticks; fresh company and configuration",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native 32bpp software framebuffer; OpenGFX 7.1",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when native company infrastructure owns at least two road bits",
          "scenario": "new 64x64 company in 1950; first owned road",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 30,
          "task_id": "openttd-first-road",
          "task_version": "1"
        },
        "mindustry-copper": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:90468c4e5dde3dfaf18e1ec5fe337b4d32906271d3a7be899efd5b87756082a2",
          "cursor_step_pixels": 32,
          "engine": "Mindustry",
          "engine_revision": "067c720a8817c1c9fb586c03898a7d948caaed56",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "160.5+ggbench1",
          "frames_per_action": 6,
          "initialization": "120 native frames; fresh player and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "optional_effects": false,
          "physics_step_seconds": 0.016666666666666666,
          "renderer": "native SDL/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "native player-plus-core copper gain from initial total; target 15",
          "scenario": "serpulo/groundZero; native first-15-copper tutorial objective",
          "score_range": [
            0,
            15
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "mindustry-copper",
          "task_version": "1"
        },
        "cdda-first-weapon": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:eee69b35296e92b6ddc84a9ffbf737ab66ab48ae42dbbc3f1ee8e236b97c1b1f",
          "control_policy": "one native key per decision; tutorial popups remain visible",
          "engine": "Cataclysm: Dark Days Ahead",
          "engine_revision": "7b2efa5cea38e4d4d97dd0e63b28b9148623da59",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.I-1+ggbench1",
          "initialization": "native tutorial world and preset character; fresh user directory",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 software / UltimateCataclysm tiles",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when the native avatar wields the tutorial baseball bat",
          "scenario": "native tutorial; first wielded baseball bat",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "cdda-first-weapon",
          "task_version": "1"
        },
        "warzone-first-derrick": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:c2c0aee72affd4e5d5be6aa03659c533e3c968365e0f62eccf1f9fbdd30f6c22",
          "cursor_step_pixels": 32,
          "engine": "Warzone 2100",
          "engine_revision": "7d3d8e51a61ceee89e6bf672a41337e1a9cd7392",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "4.7.0+ggbench1",
          "fine_cursor_step_pixels": 8,
          "initialization": "40 native frames; fresh tutorial and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL3/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native completed structure; two unit losses end task",
          "scenario": "native TUTORIAL3; first constructed oil derrick",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 100,
          "task_id": "warzone-first-derrick",
          "task_version": "1"
        }
      },
      "decisions": 336,
      "p50_ms": 8335.5658545,
      "p95_ms": 14712.849447749999,
      "max_ms": 24340.863708,
      "misses": 336,
      "latency_eligible": false,
      "latency_policy": {
        "id": "suite-p95-200-v1",
        "metric": "suite_p95_ms",
        "limit_ms": 200.0,
        "comparison": "strictly_less_than",
        "quantile_method": "linear",
        "sample_scope": "all_recorded_decisions_in_complete_suite"
      },
      "response_timeout_seconds": 120.0,
      "trust": "local-unattested",
      "official_rank": null,
      "per_game": {
        "coin-run": 100.0,
        "dodge-lanes": 100.0,
        "doom-basic": 66.19,
        "doom-defend-center": 9.524,
        "doom-defend-line": 9.524,
        "doom-take-cover": 100.0,
        "doom-health": 100.0,
        "doom-corridor": 9.091,
        "doom-home": 0.0,
        "doom-predict": 76.183,
        "procgen-bigfish": 0.0,
        "procgen-bossfight": 0.0,
        "procgen-caveflyer": 0.0,
        "procgen-chaser": 0.0,
        "procgen-climber": 0.0,
        "procgen-coinrun": 0.0,
        "procgen-dodgeball": 0.0,
        "procgen-fruitbot": 0.0,
        "procgen-heist": 0.0,
        "procgen-jumper": 0.0,
        "procgen-leaper": 0.0,
        "procgen-maze": 0.0,
        "procgen-miner": 0.0,
        "procgen-ninja": 0.0,
        "procgen-plunder": 0.0,
        "procgen-starpilot": 0.0,
        "crafter": 0.0,
        "miniworld-oneroom": 0.0,
        "pettingzoo-pistonball": 96.687,
        "atari-breakout": 0.0,
        "retro-airstriker": 0.0,
        "nethack-score": 0.0,
        "minihack-room": 0.0,
        "unity-food-collector": 0.0,
        "football-empty-goal": 0.0,
        "supertuxkart-lighthouse": 1.356,
        "luanti-chop-tree": 0.0,
        "supertux-first-coin": 0.0,
        "dcss-first-experience": 0.0,
        "openttd-first-road": 0.0,
        "mindustry-copper": 6.667,
        "cdda-first-weapon": 0.0,
        "warzone-first-derrick": 0.0
      },
      "errors": 3,
      "aborted_episodes": 3,
      "model": "gpt-6-sol",
      "provider_metadata": {
        "transport": "authenticated-cli-per-frame",
        "requested_model": "gpt-6-sol",
        "tool_events": 0,
        "observation": "png-only",
        "calls": 168,
        "prompt_version": "pixels-json-action/2"
      },
      "hardware_details": "Apple M3 Max; 48 GiB RAM; 16 CPU cores; macOS arm64. One hosted model and one game run at a time. Each action starts a fresh authenticated CLI call. CLI startup, image transfer, network time, provider inference and action validation remain in the timer. Provider hardware is not known. Normal desktop tasks can use the same Mac."
    },
    {
      "agent": "openai-gpt-5-6-sol",
      "mode": "exhibition",
      "hardware": "Darwin arm64",
      "version": "0.5.0",
      "max_steps": 8,
      "score": 15.323,
      "ci95": null,
      "episodes": 43,
      "seeds": 1,
      "seed_ids": [
        3000
      ],
      "games": [
        "coin-run",
        "dodge-lanes",
        "doom-basic",
        "doom-defend-center",
        "doom-defend-line",
        "doom-take-cover",
        "doom-health",
        "doom-corridor",
        "doom-home",
        "doom-predict",
        "procgen-bigfish",
        "procgen-bossfight",
        "procgen-caveflyer",
        "procgen-chaser",
        "procgen-climber",
        "procgen-coinrun",
        "procgen-dodgeball",
        "procgen-fruitbot",
        "procgen-heist",
        "procgen-jumper",
        "procgen-leaper",
        "procgen-maze",
        "procgen-miner",
        "procgen-ninja",
        "procgen-plunder",
        "procgen-starpilot",
        "crafter",
        "miniworld-oneroom",
        "pettingzoo-pistonball",
        "atari-breakout",
        "retro-airstriker",
        "nethack-score",
        "minihack-room",
        "unity-food-collector",
        "football-empty-goal",
        "supertuxkart-lighthouse",
        "luanti-chop-tree",
        "supertux-first-coin",
        "dcss-first-experience",
        "openttd-first-road",
        "mindustry-copper",
        "cdda-first-weapon",
        "warzone-first-derrick"
      ],
      "task_metadata": {
        "coin-run": {},
        "dodge-lanes": {},
        "doom-basic": {},
        "doom-defend-center": {},
        "doom-defend-line": {},
        "doom-take-cover": {},
        "doom-health": {},
        "doom-corridor": {},
        "doom-home": {},
        "doom-predict": {},
        "procgen-bigfish": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            40
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bigfish",
          "task_version": "1"
        },
        "procgen-bossfight": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bossfight",
          "task_version": "1"
        },
        "procgen-caveflyer": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2,
            12
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-caveflyer",
          "task_version": "1"
        },
        "procgen-chaser": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-chaser",
          "task_version": "1"
        },
        "procgen-climber": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            12.6
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-climber",
          "task_version": "1"
        },
        "procgen-coinrun": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-coinrun",
          "task_version": "1"
        },
        "procgen-dodgeball": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            19
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-dodgeball",
          "task_version": "1"
        },
        "procgen-fruitbot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -1.5,
            32.4
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-fruitbot",
          "task_version": "1"
        },
        "procgen-heist": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-heist",
          "task_version": "1"
        },
        "procgen-jumper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-jumper",
          "task_version": "1"
        },
        "procgen-leaper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-leaper",
          "task_version": "1"
        },
        "procgen-maze": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-maze",
          "task_version": "1"
        },
        "procgen-miner": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-miner",
          "task_version": "1"
        },
        "procgen-ninja": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-ninja",
          "task_version": "1"
        },
        "procgen-plunder": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            4.5,
            30
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-plunder",
          "task_version": "1"
        },
        "procgen-starpilot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2.5,
            64
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-starpilot",
          "task_version": "1"
        },
        "crafter": {
          "engine": "crafter",
          "engine_version": "1.8.3",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            22
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "crafter",
          "task_version": "2"
        },
        "miniworld-oneroom": {
          "engine": "miniworld",
          "engine_version": "2.1.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "miniworld-oneroom",
          "task_version": "1"
        },
        "pettingzoo-pistonball": {
          "engine": "pettingzoo",
          "engine_version": "1.27.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -100,
            100
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "pettingzoo-pistonball",
          "task_version": "1"
        },
        "atari-breakout": {
          "engine": "ale-py",
          "engine_version": "0.12.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            864
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "atari-breakout",
          "task_version": "1"
        },
        "retro-airstriker": {
          "engine": "stable-retro",
          "engine_version": "1.0.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            10000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "retro-airstriker",
          "task_version": "1"
        },
        "nethack-score": {
          "engine": "nle",
          "engine_version": "1.3.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "nethack-score",
          "task_version": "1"
        },
        "minihack-room": {
          "engine": "minihack",
          "engine_version": "1.0.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "minihack-room",
          "task_version": "1"
        },
        "unity-food-collector": {
          "assets_sha256": "c28e27b2d3af4391a3e1e95b0871e01a16ef657ee062bb91e7740ac9d0c99be5",
          "camera_size": [
            84,
            84
          ],
          "capture_frame_rate": 30,
          "controlled_agent": "lowest-initial-agent-id",
          "engine": "mlagents-envs",
          "engine_version": "1.1.0",
          "numpy_version": "1.23.5",
          "observation": "rendered-rgb-only",
          "peer_policy": "no-op",
          "replay": "exact-frame-and-score",
          "reward_definition": "sum of native food rewards; green +1, red -1",
          "scene": "VisualFoodCollector",
          "score_range": [
            0,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "unity-food-collector",
          "task_version": "1"
        },
        "football-empty-goal": {
          "camera_size": [
            320,
            180
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:8774faa312fa0791f01966ef3152d23c012dda5117fd01366972a0bebfc13635",
          "control_policy": "release movement and shot before each decision",
          "engine": "gfootball",
          "engine_revision": "ba9952130898bde2879b01ff2b49d3d48d26ce81",
          "engine_version": "2.10.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native goal reward, no checkpoint shaping",
          "scenario": "academy_empty_goal_close",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "football-empty-goal",
          "task_version": "1"
        },
        "supertuxkart-lighthouse": {
          "assets_sha256": "56cd754653224efde4a40aef0400a71f03f72f18d2c35a89896db052dc096b23",
          "camera_size": [
            320,
            240
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:257166494f158e6b5b3441d4b52f99b83e8cef4ff60c6ce5b333251aa7cf4738",
          "engine": "PySuperTuxKart2",
          "engine_revision": "dd70f6823f248ae1df2ce513839a9b2c8c940c39",
          "engine_variant": "deterministic-render-v2",
          "engine_version": "0.7.4+ggbrender2",
          "neutral_preroll_steps": 50,
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "Mesa software / deterministic-render-v2",
          "renderer_threads": 1,
          "renderer_vector_bits": 128,
          "replay": "exact-frame-and-score",
          "reward_definition": "max(0, native overall distance) / native track length",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_seconds": 0.2,
          "task_id": "supertuxkart-lighthouse",
          "task_version": "3",
          "track": "lighthouse"
        },
        "luanti-chop-tree": {
          "camera_size": [
            64,
            64
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:f3f2df899b8709a904350472825f3253fd1a46a8d913322634a604e389fbb0fe",
          "engine": "craftium",
          "engine_revision": "8cffe4176e793f78d00b17fa7e46ccf333a7b5b0",
          "engine_variant": "serial-lockstep-v2",
          "engine_version": "0.0.1",
          "mesh_generation_threads": 1,
          "numpy_version": "2.5.3",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native Lua tree-node dig reward; first reward completes task",
          "scenario": "Craftium/ChopTree-v0: first tree block",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "pinned world snapshot; fixed map/Lua seed on initial reset",
          "simulation": "lockstep",
          "step_seconds": 0.02,
          "task_id": "luanti-chop-tree",
          "task_version": "1",
          "weather": "disabled via upstream enable_weather setting"
        },
        "supertux-first-coin": {
          "camera_size": [
            640,
            480
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:b3f936e8da2a38c0eab38ea1acfe56c1c47829cbe1f288775a135f0a67faadbf",
          "engine": "SuperTux",
          "engine_revision": "c1ddb4f28c54de77f07aa9965231a91c0ed06708",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.6.3+ggbench1",
          "frames_per_action": 8,
          "initialization": "skip level-introduction menu and fades; fresh player/save",
          "level": "world1/welcome_antarctica.stl",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "physics_step_seconds": 0.015,
          "renderer": "SDL software",
          "replay": "exact-frame-and-score",
          "reward_definition": "native level coin count; first coin completes task",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "supertux-first-coin",
          "task_version": "1"
        },
        "dcss-first-experience": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:942c43c70e0682b3d612e7c0b6e42eb8b35aeef42aad8bb77982bbbad4d617b8",
          "control_policy": "one native key per decision; no auto-explore or auto-fight",
          "engine": "DCSS",
          "engine_revision": "19cb0921cbcb6463fcedd1bb06636446c50b1905",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.34.0+ggbench1",
          "initialization": "advance title and seed-confirmation menus before observing",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 tiles / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native experience gain; death ends task",
          "scenario": "Dungeon:1; Minotaur Fighter with war axe; first experience",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "native uint64 seed; benchmark zero maps to uint64 maximum",
          "simulation": "lockstep",
          "task_id": "dcss-first-experience",
          "task_version": "1"
        },
        "openttd-first-road": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:227f7a13d0f6a96f2574ac0b428755101ad54def2217d629d08d059f4e2996b0",
          "cursor_step_pixels": 32,
          "engine": "OpenTTD",
          "engine_revision": "14ec60f248547d4d062a1160f0fc26d742319888",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "15.3+ggbench1",
          "initialization": "16 native ticks; fresh company and configuration",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native 32bpp software framebuffer; OpenGFX 7.1",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when native company infrastructure owns at least two road bits",
          "scenario": "new 64x64 company in 1950; first owned road",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 30,
          "task_id": "openttd-first-road",
          "task_version": "1"
        },
        "mindustry-copper": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:90468c4e5dde3dfaf18e1ec5fe337b4d32906271d3a7be899efd5b87756082a2",
          "cursor_step_pixels": 32,
          "engine": "Mindustry",
          "engine_revision": "067c720a8817c1c9fb586c03898a7d948caaed56",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "160.5+ggbench1",
          "frames_per_action": 6,
          "initialization": "120 native frames; fresh player and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "optional_effects": false,
          "physics_step_seconds": 0.016666666666666666,
          "renderer": "native SDL/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "native player-plus-core copper gain from initial total; target 15",
          "scenario": "serpulo/groundZero; native first-15-copper tutorial objective",
          "score_range": [
            0,
            15
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "mindustry-copper",
          "task_version": "1"
        },
        "cdda-first-weapon": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:eee69b35296e92b6ddc84a9ffbf737ab66ab48ae42dbbc3f1ee8e236b97c1b1f",
          "control_policy": "one native key per decision; tutorial popups remain visible",
          "engine": "Cataclysm: Dark Days Ahead",
          "engine_revision": "7b2efa5cea38e4d4d97dd0e63b28b9148623da59",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.I-1+ggbench1",
          "initialization": "native tutorial world and preset character; fresh user directory",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 software / UltimateCataclysm tiles",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when the native avatar wields the tutorial baseball bat",
          "scenario": "native tutorial; first wielded baseball bat",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "cdda-first-weapon",
          "task_version": "1"
        },
        "warzone-first-derrick": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:c2c0aee72affd4e5d5be6aa03659c533e3c968365e0f62eccf1f9fbdd30f6c22",
          "cursor_step_pixels": 32,
          "engine": "Warzone 2100",
          "engine_revision": "7d3d8e51a61ceee89e6bf672a41337e1a9cd7392",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "4.7.0+ggbench1",
          "fine_cursor_step_pixels": 8,
          "initialization": "40 native frames; fresh tutorial and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL3/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native completed structure; two unit losses end task",
          "scenario": "native TUTORIAL3; first constructed oil derrick",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 100,
          "task_id": "warzone-first-derrick",
          "task_version": "1"
        }
      },
      "decisions": 344,
      "p50_ms": 9307.7145215,
      "p95_ms": 16570.744368699987,
      "max_ms": 41530.527792,
      "misses": 344,
      "latency_eligible": false,
      "latency_policy": {
        "id": "suite-p95-200-v1",
        "metric": "suite_p95_ms",
        "limit_ms": 200.0,
        "comparison": "strictly_less_than",
        "quantile_method": "linear",
        "sample_scope": "all_recorded_decisions_in_complete_suite"
      },
      "response_timeout_seconds": 120.0,
      "trust": "local-unattested",
      "official_rank": null,
      "per_game": {
        "coin-run": 100.0,
        "dodge-lanes": 100.0,
        "doom-basic": 66.19,
        "doom-defend-center": 9.524,
        "doom-defend-line": 9.524,
        "doom-take-cover": 100.0,
        "doom-health": 100.0,
        "doom-corridor": 9.091,
        "doom-home": 0.0,
        "doom-predict": 76.183,
        "procgen-bigfish": 0.0,
        "procgen-bossfight": 0.0,
        "procgen-caveflyer": 0.0,
        "procgen-chaser": 0.0,
        "procgen-climber": 0.0,
        "procgen-coinrun": 0.0,
        "procgen-dodgeball": 0.0,
        "procgen-fruitbot": 7.375,
        "procgen-heist": 0.0,
        "procgen-jumper": 0.0,
        "procgen-leaper": 0.0,
        "procgen-maze": 0.0,
        "procgen-miner": 0.0,
        "procgen-ninja": 0.0,
        "procgen-plunder": 0.0,
        "procgen-starpilot": 0.0,
        "crafter": 0.0,
        "miniworld-oneroom": 0.0,
        "pettingzoo-pistonball": 79.697,
        "atari-breakout": 0.0,
        "retro-airstriker": 0.0,
        "nethack-score": 0.0,
        "minihack-room": 0.0,
        "unity-food-collector": 0.0,
        "football-empty-goal": 0.0,
        "supertuxkart-lighthouse": 1.319,
        "luanti-chop-tree": 0.0,
        "supertux-first-coin": 0.0,
        "dcss-first-experience": 0.0,
        "openttd-first-road": 0.0,
        "mindustry-copper": 0.0,
        "cdda-first-weapon": 0.0,
        "warzone-first-derrick": 0.0
      },
      "errors": 0,
      "aborted_episodes": 0,
      "model": "gpt-5.6-sol",
      "provider_metadata": {
        "transport": "authenticated-cli-per-frame",
        "requested_model": "gpt-5.6-sol",
        "tool_events": 0,
        "observation": "png-only",
        "calls": 344,
        "prompt_version": "pixels-json-action/2"
      },
      "hardware_details": "Apple M3 Max; 48 GiB RAM; 16 CPU cores; macOS arm64. One hosted model and one game run at a time. Each action starts a fresh authenticated CLI call. CLI startup, image transfer, network time, provider inference and action validation remain in the timer. Provider hardware is not known. Normal desktop tasks can use the same Mac."
    },
    {
      "agent": "openai-gpt-6-luna",
      "mode": "exhibition",
      "hardware": "Darwin arm64",
      "version": "0.5.0",
      "max_steps": 8,
      "score": 15.03,
      "ci95": null,
      "episodes": 43,
      "seeds": 1,
      "seed_ids": [
        3000
      ],
      "games": [
        "coin-run",
        "dodge-lanes",
        "doom-basic",
        "doom-defend-center",
        "doom-defend-line",
        "doom-take-cover",
        "doom-health",
        "doom-corridor",
        "doom-home",
        "doom-predict",
        "procgen-bigfish",
        "procgen-bossfight",
        "procgen-caveflyer",
        "procgen-chaser",
        "procgen-climber",
        "procgen-coinrun",
        "procgen-dodgeball",
        "procgen-fruitbot",
        "procgen-heist",
        "procgen-jumper",
        "procgen-leaper",
        "procgen-maze",
        "procgen-miner",
        "procgen-ninja",
        "procgen-plunder",
        "procgen-starpilot",
        "crafter",
        "miniworld-oneroom",
        "pettingzoo-pistonball",
        "atari-breakout",
        "retro-airstriker",
        "nethack-score",
        "minihack-room",
        "unity-food-collector",
        "football-empty-goal",
        "supertuxkart-lighthouse",
        "luanti-chop-tree",
        "supertux-first-coin",
        "dcss-first-experience",
        "openttd-first-road",
        "mindustry-copper",
        "cdda-first-weapon",
        "warzone-first-derrick"
      ],
      "task_metadata": {
        "coin-run": {},
        "dodge-lanes": {},
        "doom-basic": {},
        "doom-defend-center": {},
        "doom-defend-line": {},
        "doom-take-cover": {},
        "doom-health": {},
        "doom-corridor": {},
        "doom-home": {},
        "doom-predict": {},
        "procgen-bigfish": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            40
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bigfish",
          "task_version": "1"
        },
        "procgen-bossfight": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bossfight",
          "task_version": "1"
        },
        "procgen-caveflyer": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2,
            12
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-caveflyer",
          "task_version": "1"
        },
        "procgen-chaser": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-chaser",
          "task_version": "1"
        },
        "procgen-climber": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            12.6
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-climber",
          "task_version": "1"
        },
        "procgen-coinrun": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-coinrun",
          "task_version": "1"
        },
        "procgen-dodgeball": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            19
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-dodgeball",
          "task_version": "1"
        },
        "procgen-fruitbot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -1.5,
            32.4
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-fruitbot",
          "task_version": "1"
        },
        "procgen-heist": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-heist",
          "task_version": "1"
        },
        "procgen-jumper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-jumper",
          "task_version": "1"
        },
        "procgen-leaper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-leaper",
          "task_version": "1"
        },
        "procgen-maze": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-maze",
          "task_version": "1"
        },
        "procgen-miner": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-miner",
          "task_version": "1"
        },
        "procgen-ninja": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-ninja",
          "task_version": "1"
        },
        "procgen-plunder": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            4.5,
            30
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-plunder",
          "task_version": "1"
        },
        "procgen-starpilot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2.5,
            64
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-starpilot",
          "task_version": "1"
        },
        "crafter": {
          "engine": "crafter",
          "engine_version": "1.8.3",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            22
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "crafter",
          "task_version": "2"
        },
        "miniworld-oneroom": {
          "engine": "miniworld",
          "engine_version": "2.1.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "miniworld-oneroom",
          "task_version": "1"
        },
        "pettingzoo-pistonball": {
          "engine": "pettingzoo",
          "engine_version": "1.27.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -100,
            100
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "pettingzoo-pistonball",
          "task_version": "1"
        },
        "atari-breakout": {
          "engine": "ale-py",
          "engine_version": "0.12.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            864
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "atari-breakout",
          "task_version": "1"
        },
        "retro-airstriker": {
          "engine": "stable-retro",
          "engine_version": "1.0.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            10000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "retro-airstriker",
          "task_version": "1"
        },
        "nethack-score": {
          "engine": "nle",
          "engine_version": "1.3.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "nethack-score",
          "task_version": "1"
        },
        "minihack-room": {
          "engine": "minihack",
          "engine_version": "1.0.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "minihack-room",
          "task_version": "1"
        },
        "unity-food-collector": {
          "assets_sha256": "c28e27b2d3af4391a3e1e95b0871e01a16ef657ee062bb91e7740ac9d0c99be5",
          "camera_size": [
            84,
            84
          ],
          "capture_frame_rate": 30,
          "controlled_agent": "lowest-initial-agent-id",
          "engine": "mlagents-envs",
          "engine_version": "1.1.0",
          "numpy_version": "1.23.5",
          "observation": "rendered-rgb-only",
          "peer_policy": "no-op",
          "replay": "exact-frame-and-score",
          "reward_definition": "sum of native food rewards; green +1, red -1",
          "scene": "VisualFoodCollector",
          "score_range": [
            0,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "unity-food-collector",
          "task_version": "1"
        },
        "football-empty-goal": {
          "camera_size": [
            320,
            180
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:8774faa312fa0791f01966ef3152d23c012dda5117fd01366972a0bebfc13635",
          "control_policy": "release movement and shot before each decision",
          "engine": "gfootball",
          "engine_revision": "ba9952130898bde2879b01ff2b49d3d48d26ce81",
          "engine_version": "2.10.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native goal reward, no checkpoint shaping",
          "scenario": "academy_empty_goal_close",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "football-empty-goal",
          "task_version": "1"
        },
        "supertuxkart-lighthouse": {
          "assets_sha256": "56cd754653224efde4a40aef0400a71f03f72f18d2c35a89896db052dc096b23",
          "camera_size": [
            320,
            240
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:257166494f158e6b5b3441d4b52f99b83e8cef4ff60c6ce5b333251aa7cf4738",
          "engine": "PySuperTuxKart2",
          "engine_revision": "dd70f6823f248ae1df2ce513839a9b2c8c940c39",
          "engine_variant": "deterministic-render-v2",
          "engine_version": "0.7.4+ggbrender2",
          "neutral_preroll_steps": 50,
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "Mesa software / deterministic-render-v2",
          "renderer_threads": 1,
          "renderer_vector_bits": 128,
          "replay": "exact-frame-and-score",
          "reward_definition": "max(0, native overall distance) / native track length",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_seconds": 0.2,
          "task_id": "supertuxkart-lighthouse",
          "task_version": "3",
          "track": "lighthouse"
        },
        "luanti-chop-tree": {
          "camera_size": [
            64,
            64
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:f3f2df899b8709a904350472825f3253fd1a46a8d913322634a604e389fbb0fe",
          "engine": "craftium",
          "engine_revision": "8cffe4176e793f78d00b17fa7e46ccf333a7b5b0",
          "engine_variant": "serial-lockstep-v2",
          "engine_version": "0.0.1",
          "mesh_generation_threads": 1,
          "numpy_version": "2.5.3",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native Lua tree-node dig reward; first reward completes task",
          "scenario": "Craftium/ChopTree-v0: first tree block",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "pinned world snapshot; fixed map/Lua seed on initial reset",
          "simulation": "lockstep",
          "step_seconds": 0.02,
          "task_id": "luanti-chop-tree",
          "task_version": "1",
          "weather": "disabled via upstream enable_weather setting"
        },
        "supertux-first-coin": {
          "camera_size": [
            640,
            480
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:b3f936e8da2a38c0eab38ea1acfe56c1c47829cbe1f288775a135f0a67faadbf",
          "engine": "SuperTux",
          "engine_revision": "c1ddb4f28c54de77f07aa9965231a91c0ed06708",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.6.3+ggbench1",
          "frames_per_action": 8,
          "initialization": "skip level-introduction menu and fades; fresh player/save",
          "level": "world1/welcome_antarctica.stl",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "physics_step_seconds": 0.015,
          "renderer": "SDL software",
          "replay": "exact-frame-and-score",
          "reward_definition": "native level coin count; first coin completes task",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "supertux-first-coin",
          "task_version": "1"
        },
        "dcss-first-experience": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:942c43c70e0682b3d612e7c0b6e42eb8b35aeef42aad8bb77982bbbad4d617b8",
          "control_policy": "one native key per decision; no auto-explore or auto-fight",
          "engine": "DCSS",
          "engine_revision": "19cb0921cbcb6463fcedd1bb06636446c50b1905",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.34.0+ggbench1",
          "initialization": "advance title and seed-confirmation menus before observing",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 tiles / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native experience gain; death ends task",
          "scenario": "Dungeon:1; Minotaur Fighter with war axe; first experience",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "native uint64 seed; benchmark zero maps to uint64 maximum",
          "simulation": "lockstep",
          "task_id": "dcss-first-experience",
          "task_version": "1"
        },
        "openttd-first-road": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:227f7a13d0f6a96f2574ac0b428755101ad54def2217d629d08d059f4e2996b0",
          "cursor_step_pixels": 32,
          "engine": "OpenTTD",
          "engine_revision": "14ec60f248547d4d062a1160f0fc26d742319888",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "15.3+ggbench1",
          "initialization": "16 native ticks; fresh company and configuration",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native 32bpp software framebuffer; OpenGFX 7.1",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when native company infrastructure owns at least two road bits",
          "scenario": "new 64x64 company in 1950; first owned road",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 30,
          "task_id": "openttd-first-road",
          "task_version": "1"
        },
        "mindustry-copper": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:90468c4e5dde3dfaf18e1ec5fe337b4d32906271d3a7be899efd5b87756082a2",
          "cursor_step_pixels": 32,
          "engine": "Mindustry",
          "engine_revision": "067c720a8817c1c9fb586c03898a7d948caaed56",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "160.5+ggbench1",
          "frames_per_action": 6,
          "initialization": "120 native frames; fresh player and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "optional_effects": false,
          "physics_step_seconds": 0.016666666666666666,
          "renderer": "native SDL/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "native player-plus-core copper gain from initial total; target 15",
          "scenario": "serpulo/groundZero; native first-15-copper tutorial objective",
          "score_range": [
            0,
            15
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "mindustry-copper",
          "task_version": "1"
        },
        "cdda-first-weapon": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:eee69b35296e92b6ddc84a9ffbf737ab66ab48ae42dbbc3f1ee8e236b97c1b1f",
          "control_policy": "one native key per decision; tutorial popups remain visible",
          "engine": "Cataclysm: Dark Days Ahead",
          "engine_revision": "7b2efa5cea38e4d4d97dd0e63b28b9148623da59",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.I-1+ggbench1",
          "initialization": "native tutorial world and preset character; fresh user directory",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 software / UltimateCataclysm tiles",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when the native avatar wields the tutorial baseball bat",
          "scenario": "native tutorial; first wielded baseball bat",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "cdda-first-weapon",
          "task_version": "1"
        },
        "warzone-first-derrick": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:c2c0aee72affd4e5d5be6aa03659c533e3c968365e0f62eccf1f9fbdd30f6c22",
          "cursor_step_pixels": 32,
          "engine": "Warzone 2100",
          "engine_revision": "7d3d8e51a61ceee89e6bf672a41337e1a9cd7392",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "4.7.0+ggbench1",
          "fine_cursor_step_pixels": 8,
          "initialization": "40 native frames; fresh tutorial and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL3/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native completed structure; two unit losses end task",
          "scenario": "native TUTORIAL3; first constructed oil derrick",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 100,
          "task_id": "warzone-first-derrick",
          "task_version": "1"
        }
      },
      "decisions": 344,
      "p50_ms": 5241.6092915,
      "p95_ms": 17991.2067731,
      "max_ms": 35733.722041,
      "misses": 344,
      "latency_eligible": false,
      "latency_policy": {
        "id": "suite-p95-200-v1",
        "metric": "suite_p95_ms",
        "limit_ms": 200.0,
        "comparison": "strictly_less_than",
        "quantile_method": "linear",
        "sample_scope": "all_recorded_decisions_in_complete_suite"
      },
      "response_timeout_seconds": 120.0,
      "trust": "local-unattested",
      "official_rank": null,
      "per_game": {
        "coin-run": 100.0,
        "dodge-lanes": 100.0,
        "doom-basic": 66.19,
        "doom-defend-center": 9.524,
        "doom-defend-line": 9.524,
        "doom-take-cover": 100.0,
        "doom-health": 100.0,
        "doom-corridor": 9.091,
        "doom-home": 0.0,
        "doom-predict": 76.183,
        "procgen-bigfish": 0.0,
        "procgen-bossfight": 0.0,
        "procgen-caveflyer": 0.0,
        "procgen-chaser": 0.0,
        "procgen-climber": 0.0,
        "procgen-coinrun": 0.0,
        "procgen-dodgeball": 0.0,
        "procgen-fruitbot": 4.425,
        "procgen-heist": 0.0,
        "procgen-jumper": 0.0,
        "procgen-leaper": 0.0,
        "procgen-maze": 0.0,
        "procgen-miner": 0.0,
        "procgen-ninja": 0.0,
        "procgen-plunder": 0.0,
        "procgen-starpilot": 0.0,
        "crafter": 0.0,
        "miniworld-oneroom": 0.0,
        "pettingzoo-pistonball": 69.988,
        "atari-breakout": 0.0,
        "retro-airstriker": 0.0,
        "nethack-score": 0.0,
        "minihack-room": 0.0,
        "unity-food-collector": 0.0,
        "football-empty-goal": 0.0,
        "supertuxkart-lighthouse": 1.362,
        "luanti-chop-tree": 0.0,
        "supertux-first-coin": 0.0,
        "dcss-first-experience": 0.0,
        "openttd-first-road": 0.0,
        "mindustry-copper": 0.0,
        "cdda-first-weapon": 0.0,
        "warzone-first-derrick": 0.0
      },
      "errors": 0,
      "aborted_episodes": 0,
      "model": "gpt-6-luna",
      "provider_metadata": {
        "transport": "authenticated-cli-per-frame",
        "requested_model": "gpt-6-luna",
        "tool_events": 0,
        "observation": "png-only",
        "calls": 344,
        "prompt_version": "pixels-json-action/2"
      },
      "hardware_details": "Apple M3 Max; 48 GiB RAM; 16 CPU cores; macOS arm64. One hosted model and one game run at a time. Each action starts a fresh authenticated CLI call. CLI startup, image transfer, network time, provider inference and action validation remain in the timer. Provider hardware is not known. Normal desktop tasks can use the same Mac."
    },
    {
      "agent": "openai-gpt-5-6-luna",
      "mode": "exhibition",
      "hardware": "Darwin arm64",
      "version": "0.5.0",
      "max_steps": 8,
      "score": 14.437,
      "ci95": null,
      "episodes": 43,
      "seeds": 1,
      "seed_ids": [
        3000
      ],
      "games": [
        "coin-run",
        "dodge-lanes",
        "doom-basic",
        "doom-defend-center",
        "doom-defend-line",
        "doom-take-cover",
        "doom-health",
        "doom-corridor",
        "doom-home",
        "doom-predict",
        "procgen-bigfish",
        "procgen-bossfight",
        "procgen-caveflyer",
        "procgen-chaser",
        "procgen-climber",
        "procgen-coinrun",
        "procgen-dodgeball",
        "procgen-fruitbot",
        "procgen-heist",
        "procgen-jumper",
        "procgen-leaper",
        "procgen-maze",
        "procgen-miner",
        "procgen-ninja",
        "procgen-plunder",
        "procgen-starpilot",
        "crafter",
        "miniworld-oneroom",
        "pettingzoo-pistonball",
        "atari-breakout",
        "retro-airstriker",
        "nethack-score",
        "minihack-room",
        "unity-food-collector",
        "football-empty-goal",
        "supertuxkart-lighthouse",
        "luanti-chop-tree",
        "supertux-first-coin",
        "dcss-first-experience",
        "openttd-first-road",
        "mindustry-copper",
        "cdda-first-weapon",
        "warzone-first-derrick"
      ],
      "task_metadata": {
        "coin-run": {},
        "dodge-lanes": {},
        "doom-basic": {},
        "doom-defend-center": {},
        "doom-defend-line": {},
        "doom-take-cover": {},
        "doom-health": {},
        "doom-corridor": {},
        "doom-home": {},
        "doom-predict": {},
        "procgen-bigfish": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            40
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bigfish",
          "task_version": "1"
        },
        "procgen-bossfight": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bossfight",
          "task_version": "1"
        },
        "procgen-caveflyer": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2,
            12
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-caveflyer",
          "task_version": "1"
        },
        "procgen-chaser": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-chaser",
          "task_version": "1"
        },
        "procgen-climber": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            12.6
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-climber",
          "task_version": "1"
        },
        "procgen-coinrun": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-coinrun",
          "task_version": "1"
        },
        "procgen-dodgeball": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            19
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-dodgeball",
          "task_version": "1"
        },
        "procgen-fruitbot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -1.5,
            32.4
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-fruitbot",
          "task_version": "1"
        },
        "procgen-heist": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-heist",
          "task_version": "1"
        },
        "procgen-jumper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-jumper",
          "task_version": "1"
        },
        "procgen-leaper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-leaper",
          "task_version": "1"
        },
        "procgen-maze": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-maze",
          "task_version": "1"
        },
        "procgen-miner": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-miner",
          "task_version": "1"
        },
        "procgen-ninja": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-ninja",
          "task_version": "1"
        },
        "procgen-plunder": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            4.5,
            30
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-plunder",
          "task_version": "1"
        },
        "procgen-starpilot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2.5,
            64
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-starpilot",
          "task_version": "1"
        },
        "crafter": {
          "engine": "crafter",
          "engine_version": "1.8.3",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            22
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "crafter",
          "task_version": "2"
        },
        "miniworld-oneroom": {
          "engine": "miniworld",
          "engine_version": "2.1.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "miniworld-oneroom",
          "task_version": "1"
        },
        "pettingzoo-pistonball": {
          "engine": "pettingzoo",
          "engine_version": "1.27.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -100,
            100
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "pettingzoo-pistonball",
          "task_version": "1"
        },
        "atari-breakout": {
          "engine": "ale-py",
          "engine_version": "0.12.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            864
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "atari-breakout",
          "task_version": "1"
        },
        "retro-airstriker": {
          "engine": "stable-retro",
          "engine_version": "1.0.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            10000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "retro-airstriker",
          "task_version": "1"
        },
        "nethack-score": {
          "engine": "nle",
          "engine_version": "1.3.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "nethack-score",
          "task_version": "1"
        },
        "minihack-room": {
          "engine": "minihack",
          "engine_version": "1.0.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "minihack-room",
          "task_version": "1"
        },
        "unity-food-collector": {
          "assets_sha256": "c28e27b2d3af4391a3e1e95b0871e01a16ef657ee062bb91e7740ac9d0c99be5",
          "camera_size": [
            84,
            84
          ],
          "capture_frame_rate": 30,
          "controlled_agent": "lowest-initial-agent-id",
          "engine": "mlagents-envs",
          "engine_version": "1.1.0",
          "numpy_version": "1.23.5",
          "observation": "rendered-rgb-only",
          "peer_policy": "no-op",
          "replay": "exact-frame-and-score",
          "reward_definition": "sum of native food rewards; green +1, red -1",
          "scene": "VisualFoodCollector",
          "score_range": [
            0,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "unity-food-collector",
          "task_version": "1"
        },
        "football-empty-goal": {
          "camera_size": [
            320,
            180
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:8774faa312fa0791f01966ef3152d23c012dda5117fd01366972a0bebfc13635",
          "control_policy": "release movement and shot before each decision",
          "engine": "gfootball",
          "engine_revision": "ba9952130898bde2879b01ff2b49d3d48d26ce81",
          "engine_version": "2.10.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native goal reward, no checkpoint shaping",
          "scenario": "academy_empty_goal_close",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "football-empty-goal",
          "task_version": "1"
        },
        "supertuxkart-lighthouse": {
          "assets_sha256": "56cd754653224efde4a40aef0400a71f03f72f18d2c35a89896db052dc096b23",
          "camera_size": [
            320,
            240
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:257166494f158e6b5b3441d4b52f99b83e8cef4ff60c6ce5b333251aa7cf4738",
          "engine": "PySuperTuxKart2",
          "engine_revision": "dd70f6823f248ae1df2ce513839a9b2c8c940c39",
          "engine_variant": "deterministic-render-v2",
          "engine_version": "0.7.4+ggbrender2",
          "neutral_preroll_steps": 50,
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "Mesa software / deterministic-render-v2",
          "renderer_threads": 1,
          "renderer_vector_bits": 128,
          "replay": "exact-frame-and-score",
          "reward_definition": "max(0, native overall distance) / native track length",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_seconds": 0.2,
          "task_id": "supertuxkart-lighthouse",
          "task_version": "3",
          "track": "lighthouse"
        },
        "luanti-chop-tree": {
          "camera_size": [
            64,
            64
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:f3f2df899b8709a904350472825f3253fd1a46a8d913322634a604e389fbb0fe",
          "engine": "craftium",
          "engine_revision": "8cffe4176e793f78d00b17fa7e46ccf333a7b5b0",
          "engine_variant": "serial-lockstep-v2",
          "engine_version": "0.0.1",
          "mesh_generation_threads": 1,
          "numpy_version": "2.5.3",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native Lua tree-node dig reward; first reward completes task",
          "scenario": "Craftium/ChopTree-v0: first tree block",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "pinned world snapshot; fixed map/Lua seed on initial reset",
          "simulation": "lockstep",
          "step_seconds": 0.02,
          "task_id": "luanti-chop-tree",
          "task_version": "1",
          "weather": "disabled via upstream enable_weather setting"
        },
        "supertux-first-coin": {
          "camera_size": [
            640,
            480
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:b3f936e8da2a38c0eab38ea1acfe56c1c47829cbe1f288775a135f0a67faadbf",
          "engine": "SuperTux",
          "engine_revision": "c1ddb4f28c54de77f07aa9965231a91c0ed06708",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.6.3+ggbench1",
          "frames_per_action": 8,
          "initialization": "skip level-introduction menu and fades; fresh player/save",
          "level": "world1/welcome_antarctica.stl",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "physics_step_seconds": 0.015,
          "renderer": "SDL software",
          "replay": "exact-frame-and-score",
          "reward_definition": "native level coin count; first coin completes task",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "supertux-first-coin",
          "task_version": "1"
        },
        "dcss-first-experience": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:942c43c70e0682b3d612e7c0b6e42eb8b35aeef42aad8bb77982bbbad4d617b8",
          "control_policy": "one native key per decision; no auto-explore or auto-fight",
          "engine": "DCSS",
          "engine_revision": "19cb0921cbcb6463fcedd1bb06636446c50b1905",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.34.0+ggbench1",
          "initialization": "advance title and seed-confirmation menus before observing",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 tiles / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native experience gain; death ends task",
          "scenario": "Dungeon:1; Minotaur Fighter with war axe; first experience",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "native uint64 seed; benchmark zero maps to uint64 maximum",
          "simulation": "lockstep",
          "task_id": "dcss-first-experience",
          "task_version": "1"
        },
        "openttd-first-road": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:227f7a13d0f6a96f2574ac0b428755101ad54def2217d629d08d059f4e2996b0",
          "cursor_step_pixels": 32,
          "engine": "OpenTTD",
          "engine_revision": "14ec60f248547d4d062a1160f0fc26d742319888",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "15.3+ggbench1",
          "initialization": "16 native ticks; fresh company and configuration",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native 32bpp software framebuffer; OpenGFX 7.1",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when native company infrastructure owns at least two road bits",
          "scenario": "new 64x64 company in 1950; first owned road",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 30,
          "task_id": "openttd-first-road",
          "task_version": "1"
        },
        "mindustry-copper": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:90468c4e5dde3dfaf18e1ec5fe337b4d32906271d3a7be899efd5b87756082a2",
          "cursor_step_pixels": 32,
          "engine": "Mindustry",
          "engine_revision": "067c720a8817c1c9fb586c03898a7d948caaed56",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "160.5+ggbench1",
          "frames_per_action": 6,
          "initialization": "120 native frames; fresh player and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "optional_effects": false,
          "physics_step_seconds": 0.016666666666666666,
          "renderer": "native SDL/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "native player-plus-core copper gain from initial total; target 15",
          "scenario": "serpulo/groundZero; native first-15-copper tutorial objective",
          "score_range": [
            0,
            15
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "mindustry-copper",
          "task_version": "1"
        },
        "cdda-first-weapon": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:eee69b35296e92b6ddc84a9ffbf737ab66ab48ae42dbbc3f1ee8e236b97c1b1f",
          "control_policy": "one native key per decision; tutorial popups remain visible",
          "engine": "Cataclysm: Dark Days Ahead",
          "engine_revision": "7b2efa5cea38e4d4d97dd0e63b28b9148623da59",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.I-1+ggbench1",
          "initialization": "native tutorial world and preset character; fresh user directory",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 software / UltimateCataclysm tiles",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when the native avatar wields the tutorial baseball bat",
          "scenario": "native tutorial; first wielded baseball bat",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "cdda-first-weapon",
          "task_version": "1"
        },
        "warzone-first-derrick": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:c2c0aee72affd4e5d5be6aa03659c533e3c968365e0f62eccf1f9fbdd30f6c22",
          "cursor_step_pixels": 32,
          "engine": "Warzone 2100",
          "engine_revision": "7d3d8e51a61ceee89e6bf672a41337e1a9cd7392",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "4.7.0+ggbench1",
          "fine_cursor_step_pixels": 8,
          "initialization": "40 native frames; fresh tutorial and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL3/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native completed structure; two unit losses end task",
          "scenario": "native TUTORIAL3; first constructed oil derrick",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 100,
          "task_id": "warzone-first-derrick",
          "task_version": "1"
        }
      },
      "decisions": 344,
      "p50_ms": 7016.484270999999,
      "p95_ms": 13674.887067049998,
      "max_ms": 45417.729667,
      "misses": 344,
      "latency_eligible": false,
      "latency_policy": {
        "id": "suite-p95-200-v1",
        "metric": "suite_p95_ms",
        "limit_ms": 200.0,
        "comparison": "strictly_less_than",
        "quantile_method": "linear",
        "sample_scope": "all_recorded_decisions_in_complete_suite"
      },
      "response_timeout_seconds": 120.0,
      "trust": "local-unattested",
      "official_rank": null,
      "per_game": {
        "coin-run": 75.0,
        "dodge-lanes": 100.0,
        "doom-basic": 66.19,
        "doom-defend-center": 9.524,
        "doom-defend-line": 9.524,
        "doom-take-cover": 100.0,
        "doom-health": 100.0,
        "doom-corridor": 9.091,
        "doom-home": 0.0,
        "doom-predict": 76.183,
        "procgen-bigfish": 0.0,
        "procgen-bossfight": 0.0,
        "procgen-caveflyer": 0.0,
        "procgen-chaser": 0.0,
        "procgen-climber": 0.0,
        "procgen-coinrun": 0.0,
        "procgen-dodgeball": 0.0,
        "procgen-fruitbot": 4.425,
        "procgen-heist": 0.0,
        "procgen-jumper": 0.0,
        "procgen-leaper": 0.0,
        "procgen-maze": 0.0,
        "procgen-miner": 0.0,
        "procgen-ninja": 0.0,
        "procgen-plunder": 0.0,
        "procgen-starpilot": 0.0,
        "crafter": 0.0,
        "miniworld-oneroom": 0.0,
        "pettingzoo-pistonball": 69.503,
        "atari-breakout": 0.0,
        "retro-airstriker": 0.0,
        "nethack-score": 0.0,
        "minihack-room": 0.0,
        "unity-food-collector": 0.0,
        "football-empty-goal": 0.0,
        "supertuxkart-lighthouse": 1.337,
        "luanti-chop-tree": 0.0,
        "supertux-first-coin": 0.0,
        "dcss-first-experience": 0.0,
        "openttd-first-road": 0.0,
        "mindustry-copper": 0.0,
        "cdda-first-weapon": 0.0,
        "warzone-first-derrick": 0.0
      },
      "errors": 0,
      "aborted_episodes": 0,
      "model": "gpt-5.6-luna",
      "provider_metadata": {
        "transport": "authenticated-cli-per-frame",
        "requested_model": "gpt-5.6-luna",
        "tool_events": 0,
        "observation": "png-only",
        "calls": 344,
        "prompt_version": "pixels-json-action/2"
      },
      "hardware_details": "Apple M3 Max; 48 GiB RAM; 16 CPU cores; macOS arm64. One hosted model and one game run at a time. Each action starts a fresh authenticated CLI call. CLI startup, image transfer, network time, provider inference and action validation remain in the timer. Provider hardware is not known. Normal desktop tasks can use the same Mac."
    },
    {
      "agent": "claude-claude-sonnet-5-5",
      "mode": "exhibition",
      "hardware": "Darwin arm64",
      "version": "0.5.0",
      "max_steps": 8,
      "score": 14.176,
      "ci95": null,
      "episodes": 43,
      "seeds": 1,
      "seed_ids": [
        3000
      ],
      "games": [
        "coin-run",
        "dodge-lanes",
        "doom-basic",
        "doom-defend-center",
        "doom-defend-line",
        "doom-take-cover",
        "doom-health",
        "doom-corridor",
        "doom-home",
        "doom-predict",
        "procgen-bigfish",
        "procgen-bossfight",
        "procgen-caveflyer",
        "procgen-chaser",
        "procgen-climber",
        "procgen-coinrun",
        "procgen-dodgeball",
        "procgen-fruitbot",
        "procgen-heist",
        "procgen-jumper",
        "procgen-leaper",
        "procgen-maze",
        "procgen-miner",
        "procgen-ninja",
        "procgen-plunder",
        "procgen-starpilot",
        "crafter",
        "miniworld-oneroom",
        "pettingzoo-pistonball",
        "atari-breakout",
        "retro-airstriker",
        "nethack-score",
        "minihack-room",
        "unity-food-collector",
        "football-empty-goal",
        "supertuxkart-lighthouse",
        "luanti-chop-tree",
        "supertux-first-coin",
        "dcss-first-experience",
        "openttd-first-road",
        "mindustry-copper",
        "cdda-first-weapon",
        "warzone-first-derrick"
      ],
      "task_metadata": {
        "coin-run": {},
        "dodge-lanes": {},
        "doom-basic": {},
        "doom-defend-center": {},
        "doom-defend-line": {},
        "doom-take-cover": {},
        "doom-health": {},
        "doom-corridor": {},
        "doom-home": {},
        "doom-predict": {},
        "procgen-bigfish": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            40
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bigfish",
          "task_version": "1"
        },
        "procgen-bossfight": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bossfight",
          "task_version": "1"
        },
        "procgen-caveflyer": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2,
            12
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-caveflyer",
          "task_version": "1"
        },
        "procgen-chaser": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-chaser",
          "task_version": "1"
        },
        "procgen-climber": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            12.6
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-climber",
          "task_version": "1"
        },
        "procgen-coinrun": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-coinrun",
          "task_version": "1"
        },
        "procgen-dodgeball": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            19
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-dodgeball",
          "task_version": "1"
        },
        "procgen-fruitbot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -1.5,
            32.4
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-fruitbot",
          "task_version": "1"
        },
        "procgen-heist": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-heist",
          "task_version": "1"
        },
        "procgen-jumper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-jumper",
          "task_version": "1"
        },
        "procgen-leaper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-leaper",
          "task_version": "1"
        },
        "procgen-maze": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-maze",
          "task_version": "1"
        },
        "procgen-miner": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-miner",
          "task_version": "1"
        },
        "procgen-ninja": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-ninja",
          "task_version": "1"
        },
        "procgen-plunder": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            4.5,
            30
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-plunder",
          "task_version": "1"
        },
        "procgen-starpilot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2.5,
            64
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-starpilot",
          "task_version": "1"
        },
        "crafter": {
          "engine": "crafter",
          "engine_version": "1.8.3",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            22
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "crafter",
          "task_version": "2"
        },
        "miniworld-oneroom": {
          "engine": "miniworld",
          "engine_version": "2.1.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "miniworld-oneroom",
          "task_version": "1"
        },
        "pettingzoo-pistonball": {
          "engine": "pettingzoo",
          "engine_version": "1.27.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -100,
            100
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "pettingzoo-pistonball",
          "task_version": "1"
        },
        "atari-breakout": {
          "engine": "ale-py",
          "engine_version": "0.12.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            864
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "atari-breakout",
          "task_version": "1"
        },
        "retro-airstriker": {
          "engine": "stable-retro",
          "engine_version": "1.0.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            10000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "retro-airstriker",
          "task_version": "1"
        },
        "nethack-score": {
          "engine": "nle",
          "engine_version": "1.3.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "nethack-score",
          "task_version": "1"
        },
        "minihack-room": {
          "engine": "minihack",
          "engine_version": "1.0.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "minihack-room",
          "task_version": "1"
        },
        "unity-food-collector": {
          "assets_sha256": "c28e27b2d3af4391a3e1e95b0871e01a16ef657ee062bb91e7740ac9d0c99be5",
          "camera_size": [
            84,
            84
          ],
          "capture_frame_rate": 30,
          "controlled_agent": "lowest-initial-agent-id",
          "engine": "mlagents-envs",
          "engine_version": "1.1.0",
          "numpy_version": "1.23.5",
          "observation": "rendered-rgb-only",
          "peer_policy": "no-op",
          "replay": "exact-frame-and-score",
          "reward_definition": "sum of native food rewards; green +1, red -1",
          "scene": "VisualFoodCollector",
          "score_range": [
            0,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "unity-food-collector",
          "task_version": "1"
        },
        "football-empty-goal": {
          "camera_size": [
            320,
            180
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:8774faa312fa0791f01966ef3152d23c012dda5117fd01366972a0bebfc13635",
          "control_policy": "release movement and shot before each decision",
          "engine": "gfootball",
          "engine_revision": "ba9952130898bde2879b01ff2b49d3d48d26ce81",
          "engine_version": "2.10.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native goal reward, no checkpoint shaping",
          "scenario": "academy_empty_goal_close",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "football-empty-goal",
          "task_version": "1"
        },
        "supertuxkart-lighthouse": {
          "assets_sha256": "56cd754653224efde4a40aef0400a71f03f72f18d2c35a89896db052dc096b23",
          "camera_size": [
            320,
            240
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:257166494f158e6b5b3441d4b52f99b83e8cef4ff60c6ce5b333251aa7cf4738",
          "engine": "PySuperTuxKart2",
          "engine_revision": "dd70f6823f248ae1df2ce513839a9b2c8c940c39",
          "engine_variant": "deterministic-render-v2",
          "engine_version": "0.7.4+ggbrender2",
          "neutral_preroll_steps": 50,
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "Mesa software / deterministic-render-v2",
          "renderer_threads": 1,
          "renderer_vector_bits": 128,
          "replay": "exact-frame-and-score",
          "reward_definition": "max(0, native overall distance) / native track length",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_seconds": 0.2,
          "task_id": "supertuxkart-lighthouse",
          "task_version": "3",
          "track": "lighthouse"
        },
        "luanti-chop-tree": {
          "camera_size": [
            64,
            64
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:f3f2df899b8709a904350472825f3253fd1a46a8d913322634a604e389fbb0fe",
          "engine": "craftium",
          "engine_revision": "8cffe4176e793f78d00b17fa7e46ccf333a7b5b0",
          "engine_variant": "serial-lockstep-v2",
          "engine_version": "0.0.1",
          "mesh_generation_threads": 1,
          "numpy_version": "2.5.3",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native Lua tree-node dig reward; first reward completes task",
          "scenario": "Craftium/ChopTree-v0: first tree block",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "pinned world snapshot; fixed map/Lua seed on initial reset",
          "simulation": "lockstep",
          "step_seconds": 0.02,
          "task_id": "luanti-chop-tree",
          "task_version": "1",
          "weather": "disabled via upstream enable_weather setting"
        },
        "supertux-first-coin": {
          "camera_size": [
            640,
            480
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:b3f936e8da2a38c0eab38ea1acfe56c1c47829cbe1f288775a135f0a67faadbf",
          "engine": "SuperTux",
          "engine_revision": "c1ddb4f28c54de77f07aa9965231a91c0ed06708",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.6.3+ggbench1",
          "frames_per_action": 8,
          "initialization": "skip level-introduction menu and fades; fresh player/save",
          "level": "world1/welcome_antarctica.stl",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "physics_step_seconds": 0.015,
          "renderer": "SDL software",
          "replay": "exact-frame-and-score",
          "reward_definition": "native level coin count; first coin completes task",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "supertux-first-coin",
          "task_version": "1"
        },
        "dcss-first-experience": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:942c43c70e0682b3d612e7c0b6e42eb8b35aeef42aad8bb77982bbbad4d617b8",
          "control_policy": "one native key per decision; no auto-explore or auto-fight",
          "engine": "DCSS",
          "engine_revision": "19cb0921cbcb6463fcedd1bb06636446c50b1905",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.34.0+ggbench1",
          "initialization": "advance title and seed-confirmation menus before observing",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 tiles / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native experience gain; death ends task",
          "scenario": "Dungeon:1; Minotaur Fighter with war axe; first experience",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "native uint64 seed; benchmark zero maps to uint64 maximum",
          "simulation": "lockstep",
          "task_id": "dcss-first-experience",
          "task_version": "1"
        },
        "openttd-first-road": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:227f7a13d0f6a96f2574ac0b428755101ad54def2217d629d08d059f4e2996b0",
          "cursor_step_pixels": 32,
          "engine": "OpenTTD",
          "engine_revision": "14ec60f248547d4d062a1160f0fc26d742319888",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "15.3+ggbench1",
          "initialization": "16 native ticks; fresh company and configuration",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native 32bpp software framebuffer; OpenGFX 7.1",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when native company infrastructure owns at least two road bits",
          "scenario": "new 64x64 company in 1950; first owned road",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 30,
          "task_id": "openttd-first-road",
          "task_version": "1"
        },
        "mindustry-copper": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:90468c4e5dde3dfaf18e1ec5fe337b4d32906271d3a7be899efd5b87756082a2",
          "cursor_step_pixels": 32,
          "engine": "Mindustry",
          "engine_revision": "067c720a8817c1c9fb586c03898a7d948caaed56",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "160.5+ggbench1",
          "frames_per_action": 6,
          "initialization": "120 native frames; fresh player and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "optional_effects": false,
          "physics_step_seconds": 0.016666666666666666,
          "renderer": "native SDL/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "native player-plus-core copper gain from initial total; target 15",
          "scenario": "serpulo/groundZero; native first-15-copper tutorial objective",
          "score_range": [
            0,
            15
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "mindustry-copper",
          "task_version": "1"
        },
        "cdda-first-weapon": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:eee69b35296e92b6ddc84a9ffbf737ab66ab48ae42dbbc3f1ee8e236b97c1b1f",
          "control_policy": "one native key per decision; tutorial popups remain visible",
          "engine": "Cataclysm: Dark Days Ahead",
          "engine_revision": "7b2efa5cea38e4d4d97dd0e63b28b9148623da59",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.I-1+ggbench1",
          "initialization": "native tutorial world and preset character; fresh user directory",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 software / UltimateCataclysm tiles",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when the native avatar wields the tutorial baseball bat",
          "scenario": "native tutorial; first wielded baseball bat",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "cdda-first-weapon",
          "task_version": "1"
        },
        "warzone-first-derrick": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:c2c0aee72affd4e5d5be6aa03659c533e3c968365e0f62eccf1f9fbdd30f6c22",
          "cursor_step_pixels": 32,
          "engine": "Warzone 2100",
          "engine_revision": "7d3d8e51a61ceee89e6bf672a41337e1a9cd7392",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "4.7.0+ggbench1",
          "fine_cursor_step_pixels": 8,
          "initialization": "40 native frames; fresh tutorial and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL3/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native completed structure; two unit losses end task",
          "scenario": "native TUTORIAL3; first constructed oil derrick",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 100,
          "task_id": "warzone-first-derrick",
          "task_version": "1"
        }
      },
      "decisions": 320,
      "p50_ms": 2632.311,
      "p95_ms": 4268.193458300002,
      "max_ms": 9619.124583,
      "misses": 320,
      "latency_eligible": false,
      "latency_policy": {
        "id": "suite-p95-200-v1",
        "metric": "suite_p95_ms",
        "limit_ms": 200.0,
        "comparison": "strictly_less_than",
        "quantile_method": "linear",
        "sample_scope": "all_recorded_decisions_in_complete_suite"
      },
      "response_timeout_seconds": 120.0,
      "trust": "local-unattested",
      "official_rank": null,
      "per_game": {
        "coin-run": 0.0,
        "dodge-lanes": 100.0,
        "doom-basic": 0.0,
        "doom-defend-center": 9.524,
        "doom-defend-line": 9.524,
        "doom-take-cover": 100.0,
        "doom-health": 100.0,
        "doom-corridor": 9.091,
        "doom-home": 0.0,
        "doom-predict": 76.183,
        "procgen-bigfish": 0.0,
        "procgen-bossfight": 0.0,
        "procgen-caveflyer": 0.0,
        "procgen-chaser": 0.0,
        "procgen-climber": 0.0,
        "procgen-coinrun": 0.0,
        "procgen-dodgeball": 0.0,
        "procgen-fruitbot": 4.425,
        "procgen-heist": 0.0,
        "procgen-jumper": 0.0,
        "procgen-leaper": 0.0,
        "procgen-maze": 0.0,
        "procgen-miner": 0.0,
        "procgen-ninja": 0.0,
        "procgen-plunder": 0.0,
        "procgen-starpilot": 0.0,
        "crafter": 0.0,
        "miniworld-oneroom": 0.0,
        "pettingzoo-pistonball": 92.804,
        "atari-breakout": 0.0,
        "retro-airstriker": 0.0,
        "nethack-score": 0.0,
        "minihack-room": 100.0,
        "unity-food-collector": 0.0,
        "football-empty-goal": 0.0,
        "supertuxkart-lighthouse": 1.362,
        "luanti-chop-tree": 0.0,
        "supertux-first-coin": 0.0,
        "dcss-first-experience": 0.0,
        "openttd-first-road": 0.0,
        "mindustry-copper": 6.667,
        "cdda-first-weapon": 0.0,
        "warzone-first-derrick": 0.0
      },
      "errors": 4,
      "aborted_episodes": 4,
      "model": "claude-sonnet-5-5",
      "provider_metadata": {
        "transport": "authenticated-cli-per-frame",
        "requested_model": "claude-sonnet-5-5",
        "tool_events": 0,
        "observation": "png-only",
        "calls": 8,
        "prompt_version": "pixels-json-action/2",
        "resolved_models": [
          "claude-sonnet-5-5"
        ]
      },
      "hardware_details": "Apple M3 Max; 48 GiB RAM; 16 CPU cores; macOS arm64. One hosted model and one game run at a time. Each action starts a fresh authenticated CLI call. CLI startup, image transfer, network time, provider inference and action validation remain in the timer. Provider hardware is not known. Normal desktop tasks can use the same Mac."
    },
    {
      "agent": "claude-claude-haiku-4-5-20251001",
      "mode": "exhibition",
      "hardware": "Darwin arm64",
      "version": "0.5.0",
      "max_steps": 8,
      "score": 13.201,
      "ci95": null,
      "episodes": 43,
      "seeds": 1,
      "seed_ids": [
        3000
      ],
      "games": [
        "coin-run",
        "dodge-lanes",
        "doom-basic",
        "doom-defend-center",
        "doom-defend-line",
        "doom-take-cover",
        "doom-health",
        "doom-corridor",
        "doom-home",
        "doom-predict",
        "procgen-bigfish",
        "procgen-bossfight",
        "procgen-caveflyer",
        "procgen-chaser",
        "procgen-climber",
        "procgen-coinrun",
        "procgen-dodgeball",
        "procgen-fruitbot",
        "procgen-heist",
        "procgen-jumper",
        "procgen-leaper",
        "procgen-maze",
        "procgen-miner",
        "procgen-ninja",
        "procgen-plunder",
        "procgen-starpilot",
        "crafter",
        "miniworld-oneroom",
        "pettingzoo-pistonball",
        "atari-breakout",
        "retro-airstriker",
        "nethack-score",
        "minihack-room",
        "unity-food-collector",
        "football-empty-goal",
        "supertuxkart-lighthouse",
        "luanti-chop-tree",
        "supertux-first-coin",
        "dcss-first-experience",
        "openttd-first-road",
        "mindustry-copper",
        "cdda-first-weapon",
        "warzone-first-derrick"
      ],
      "task_metadata": {
        "coin-run": {},
        "dodge-lanes": {},
        "doom-basic": {},
        "doom-defend-center": {},
        "doom-defend-line": {},
        "doom-take-cover": {},
        "doom-health": {},
        "doom-corridor": {},
        "doom-home": {},
        "doom-predict": {},
        "procgen-bigfish": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            40
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bigfish",
          "task_version": "1"
        },
        "procgen-bossfight": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bossfight",
          "task_version": "1"
        },
        "procgen-caveflyer": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2,
            12
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-caveflyer",
          "task_version": "1"
        },
        "procgen-chaser": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-chaser",
          "task_version": "1"
        },
        "procgen-climber": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            12.6
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-climber",
          "task_version": "1"
        },
        "procgen-coinrun": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-coinrun",
          "task_version": "1"
        },
        "procgen-dodgeball": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            19
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-dodgeball",
          "task_version": "1"
        },
        "procgen-fruitbot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -1.5,
            32.4
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-fruitbot",
          "task_version": "1"
        },
        "procgen-heist": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-heist",
          "task_version": "1"
        },
        "procgen-jumper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-jumper",
          "task_version": "1"
        },
        "procgen-leaper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-leaper",
          "task_version": "1"
        },
        "procgen-maze": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-maze",
          "task_version": "1"
        },
        "procgen-miner": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-miner",
          "task_version": "1"
        },
        "procgen-ninja": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-ninja",
          "task_version": "1"
        },
        "procgen-plunder": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            4.5,
            30
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-plunder",
          "task_version": "1"
        },
        "procgen-starpilot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2.5,
            64
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-starpilot",
          "task_version": "1"
        },
        "crafter": {
          "engine": "crafter",
          "engine_version": "1.8.3",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            22
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "crafter",
          "task_version": "2"
        },
        "miniworld-oneroom": {
          "engine": "miniworld",
          "engine_version": "2.1.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "miniworld-oneroom",
          "task_version": "1"
        },
        "pettingzoo-pistonball": {
          "engine": "pettingzoo",
          "engine_version": "1.27.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -100,
            100
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "pettingzoo-pistonball",
          "task_version": "1"
        },
        "atari-breakout": {
          "engine": "ale-py",
          "engine_version": "0.12.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            864
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "atari-breakout",
          "task_version": "1"
        },
        "retro-airstriker": {
          "engine": "stable-retro",
          "engine_version": "1.0.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            10000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "retro-airstriker",
          "task_version": "1"
        },
        "nethack-score": {
          "engine": "nle",
          "engine_version": "1.3.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "nethack-score",
          "task_version": "1"
        },
        "minihack-room": {
          "engine": "minihack",
          "engine_version": "1.0.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "minihack-room",
          "task_version": "1"
        },
        "unity-food-collector": {
          "assets_sha256": "c28e27b2d3af4391a3e1e95b0871e01a16ef657ee062bb91e7740ac9d0c99be5",
          "camera_size": [
            84,
            84
          ],
          "capture_frame_rate": 30,
          "controlled_agent": "lowest-initial-agent-id",
          "engine": "mlagents-envs",
          "engine_version": "1.1.0",
          "numpy_version": "1.23.5",
          "observation": "rendered-rgb-only",
          "peer_policy": "no-op",
          "replay": "exact-frame-and-score",
          "reward_definition": "sum of native food rewards; green +1, red -1",
          "scene": "VisualFoodCollector",
          "score_range": [
            0,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "unity-food-collector",
          "task_version": "1"
        },
        "football-empty-goal": {
          "camera_size": [
            320,
            180
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:8774faa312fa0791f01966ef3152d23c012dda5117fd01366972a0bebfc13635",
          "control_policy": "release movement and shot before each decision",
          "engine": "gfootball",
          "engine_revision": "ba9952130898bde2879b01ff2b49d3d48d26ce81",
          "engine_version": "2.10.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native goal reward, no checkpoint shaping",
          "scenario": "academy_empty_goal_close",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "football-empty-goal",
          "task_version": "1"
        },
        "supertuxkart-lighthouse": {
          "assets_sha256": "56cd754653224efde4a40aef0400a71f03f72f18d2c35a89896db052dc096b23",
          "camera_size": [
            320,
            240
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:257166494f158e6b5b3441d4b52f99b83e8cef4ff60c6ce5b333251aa7cf4738",
          "engine": "PySuperTuxKart2",
          "engine_revision": "dd70f6823f248ae1df2ce513839a9b2c8c940c39",
          "engine_variant": "deterministic-render-v2",
          "engine_version": "0.7.4+ggbrender2",
          "neutral_preroll_steps": 50,
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "Mesa software / deterministic-render-v2",
          "renderer_threads": 1,
          "renderer_vector_bits": 128,
          "replay": "exact-frame-and-score",
          "reward_definition": "max(0, native overall distance) / native track length",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_seconds": 0.2,
          "task_id": "supertuxkart-lighthouse",
          "task_version": "3",
          "track": "lighthouse"
        },
        "luanti-chop-tree": {
          "camera_size": [
            64,
            64
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:f3f2df899b8709a904350472825f3253fd1a46a8d913322634a604e389fbb0fe",
          "engine": "craftium",
          "engine_revision": "8cffe4176e793f78d00b17fa7e46ccf333a7b5b0",
          "engine_variant": "serial-lockstep-v2",
          "engine_version": "0.0.1",
          "mesh_generation_threads": 1,
          "numpy_version": "2.5.3",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native Lua tree-node dig reward; first reward completes task",
          "scenario": "Craftium/ChopTree-v0: first tree block",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "pinned world snapshot; fixed map/Lua seed on initial reset",
          "simulation": "lockstep",
          "step_seconds": 0.02,
          "task_id": "luanti-chop-tree",
          "task_version": "1",
          "weather": "disabled via upstream enable_weather setting"
        },
        "supertux-first-coin": {
          "camera_size": [
            640,
            480
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:b3f936e8da2a38c0eab38ea1acfe56c1c47829cbe1f288775a135f0a67faadbf",
          "engine": "SuperTux",
          "engine_revision": "c1ddb4f28c54de77f07aa9965231a91c0ed06708",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.6.3+ggbench1",
          "frames_per_action": 8,
          "initialization": "skip level-introduction menu and fades; fresh player/save",
          "level": "world1/welcome_antarctica.stl",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "physics_step_seconds": 0.015,
          "renderer": "SDL software",
          "replay": "exact-frame-and-score",
          "reward_definition": "native level coin count; first coin completes task",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "supertux-first-coin",
          "task_version": "1"
        },
        "dcss-first-experience": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:942c43c70e0682b3d612e7c0b6e42eb8b35aeef42aad8bb77982bbbad4d617b8",
          "control_policy": "one native key per decision; no auto-explore or auto-fight",
          "engine": "DCSS",
          "engine_revision": "19cb0921cbcb6463fcedd1bb06636446c50b1905",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.34.0+ggbench1",
          "initialization": "advance title and seed-confirmation menus before observing",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 tiles / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native experience gain; death ends task",
          "scenario": "Dungeon:1; Minotaur Fighter with war axe; first experience",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "native uint64 seed; benchmark zero maps to uint64 maximum",
          "simulation": "lockstep",
          "task_id": "dcss-first-experience",
          "task_version": "1"
        },
        "openttd-first-road": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:227f7a13d0f6a96f2574ac0b428755101ad54def2217d629d08d059f4e2996b0",
          "cursor_step_pixels": 32,
          "engine": "OpenTTD",
          "engine_revision": "14ec60f248547d4d062a1160f0fc26d742319888",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "15.3+ggbench1",
          "initialization": "16 native ticks; fresh company and configuration",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native 32bpp software framebuffer; OpenGFX 7.1",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when native company infrastructure owns at least two road bits",
          "scenario": "new 64x64 company in 1950; first owned road",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 30,
          "task_id": "openttd-first-road",
          "task_version": "1"
        },
        "mindustry-copper": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:90468c4e5dde3dfaf18e1ec5fe337b4d32906271d3a7be899efd5b87756082a2",
          "cursor_step_pixels": 32,
          "engine": "Mindustry",
          "engine_revision": "067c720a8817c1c9fb586c03898a7d948caaed56",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "160.5+ggbench1",
          "frames_per_action": 6,
          "initialization": "120 native frames; fresh player and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "optional_effects": false,
          "physics_step_seconds": 0.016666666666666666,
          "renderer": "native SDL/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "native player-plus-core copper gain from initial total; target 15",
          "scenario": "serpulo/groundZero; native first-15-copper tutorial objective",
          "score_range": [
            0,
            15
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "mindustry-copper",
          "task_version": "1"
        },
        "cdda-first-weapon": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:eee69b35296e92b6ddc84a9ffbf737ab66ab48ae42dbbc3f1ee8e236b97c1b1f",
          "control_policy": "one native key per decision; tutorial popups remain visible",
          "engine": "Cataclysm: Dark Days Ahead",
          "engine_revision": "7b2efa5cea38e4d4d97dd0e63b28b9148623da59",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.I-1+ggbench1",
          "initialization": "native tutorial world and preset character; fresh user directory",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 software / UltimateCataclysm tiles",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when the native avatar wields the tutorial baseball bat",
          "scenario": "native tutorial; first wielded baseball bat",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "cdda-first-weapon",
          "task_version": "1"
        },
        "warzone-first-derrick": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:c2c0aee72affd4e5d5be6aa03659c533e3c968365e0f62eccf1f9fbdd30f6c22",
          "cursor_step_pixels": 32,
          "engine": "Warzone 2100",
          "engine_revision": "7d3d8e51a61ceee89e6bf672a41337e1a9cd7392",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "4.7.0+ggbench1",
          "fine_cursor_step_pixels": 8,
          "initialization": "40 native frames; fresh tutorial and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL3/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native completed structure; two unit losses end task",
          "scenario": "native TUTORIAL3; first constructed oil derrick",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 100,
          "task_id": "warzone-first-derrick",
          "task_version": "1"
        }
      },
      "decisions": 331,
      "p50_ms": 6454.346292,
      "p95_ms": 15912.665583,
      "max_ms": 27947.9255,
      "misses": 331,
      "latency_eligible": false,
      "latency_policy": {
        "id": "suite-p95-200-v1",
        "metric": "suite_p95_ms",
        "limit_ms": 200.0,
        "comparison": "strictly_less_than",
        "quantile_method": "linear",
        "sample_scope": "all_recorded_decisions_in_complete_suite"
      },
      "response_timeout_seconds": 120.0,
      "trust": "local-unattested",
      "official_rank": null,
      "per_game": {
        "coin-run": 0.0,
        "dodge-lanes": 100.0,
        "doom-basic": 66.19,
        "doom-defend-center": 9.524,
        "doom-defend-line": 9.524,
        "doom-take-cover": 100.0,
        "doom-health": 100.0,
        "doom-corridor": 9.091,
        "doom-home": 0.0,
        "doom-predict": 76.183,
        "procgen-bigfish": 0.0,
        "procgen-bossfight": 0.0,
        "procgen-caveflyer": 0.0,
        "procgen-chaser": 0.0,
        "procgen-climber": 0.0,
        "procgen-coinrun": 0.0,
        "procgen-dodgeball": 0.0,
        "procgen-fruitbot": 4.425,
        "procgen-heist": 0.0,
        "procgen-jumper": 0.0,
        "procgen-leaper": 0.0,
        "procgen-maze": 0.0,
        "procgen-miner": 0.0,
        "procgen-ninja": 0.0,
        "procgen-plunder": 0.0,
        "procgen-starpilot": 0.0,
        "crafter": 0.0,
        "miniworld-oneroom": 0.0,
        "pettingzoo-pistonball": 91.348,
        "atari-breakout": 0.0,
        "retro-airstriker": 0.0,
        "nethack-score": 0.0,
        "minihack-room": 0.0,
        "unity-food-collector": 0.0,
        "football-empty-goal": 0.0,
        "supertuxkart-lighthouse": 1.348,
        "luanti-chop-tree": 0.0,
        "supertux-first-coin": 0.0,
        "dcss-first-experience": 0.0,
        "openttd-first-road": 0.0,
        "mindustry-copper": 0.0,
        "cdda-first-weapon": 0.0,
        "warzone-first-derrick": 0.0
      },
      "errors": 4,
      "aborted_episodes": 4,
      "model": "claude-haiku-4-5-20251001",
      "provider_metadata": {
        "transport": "authenticated-cli-per-frame",
        "requested_model": "claude-haiku-4-5-20251001",
        "tool_events": 0,
        "observation": "png-only",
        "calls": 136,
        "prompt_version": "pixels-json-action/2",
        "resolved_models": [
          "claude-haiku-4-5-20251001"
        ]
      },
      "hardware_details": "Apple M3 Max; 48 GiB RAM; 16 CPU cores; macOS arm64. One hosted model and one game run at a time. Each action starts a fresh authenticated CLI call. CLI startup, image transfer, network time, provider inference and action validation remain in the timer. Provider hardware is not known. Normal desktop tasks can use the same Mac."
    },
    {
      "agent": "mlx-LiquidAI-LFM2-5-VL-450M-MLX-4bit",
      "mode": "exhibition",
      "hardware": "Darwin arm64",
      "version": "0.5.0",
      "max_steps": 8,
      "score": 12.718,
      "ci95": null,
      "episodes": 43,
      "seeds": 1,
      "seed_ids": [
        3000
      ],
      "games": [
        "coin-run",
        "dodge-lanes",
        "doom-basic",
        "doom-defend-center",
        "doom-defend-line",
        "doom-take-cover",
        "doom-health",
        "doom-corridor",
        "doom-home",
        "doom-predict",
        "procgen-bigfish",
        "procgen-bossfight",
        "procgen-caveflyer",
        "procgen-chaser",
        "procgen-climber",
        "procgen-coinrun",
        "procgen-dodgeball",
        "procgen-fruitbot",
        "procgen-heist",
        "procgen-jumper",
        "procgen-leaper",
        "procgen-maze",
        "procgen-miner",
        "procgen-ninja",
        "procgen-plunder",
        "procgen-starpilot",
        "crafter",
        "miniworld-oneroom",
        "pettingzoo-pistonball",
        "atari-breakout",
        "retro-airstriker",
        "nethack-score",
        "minihack-room",
        "unity-food-collector",
        "football-empty-goal",
        "supertuxkart-lighthouse",
        "luanti-chop-tree",
        "supertux-first-coin",
        "dcss-first-experience",
        "openttd-first-road",
        "mindustry-copper",
        "cdda-first-weapon",
        "warzone-first-derrick"
      ],
      "task_metadata": {
        "coin-run": {},
        "dodge-lanes": {},
        "doom-basic": {},
        "doom-defend-center": {},
        "doom-defend-line": {},
        "doom-take-cover": {},
        "doom-health": {},
        "doom-corridor": {},
        "doom-home": {},
        "doom-predict": {},
        "procgen-bigfish": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            40
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bigfish",
          "task_version": "1"
        },
        "procgen-bossfight": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bossfight",
          "task_version": "1"
        },
        "procgen-caveflyer": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2,
            12
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-caveflyer",
          "task_version": "1"
        },
        "procgen-chaser": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-chaser",
          "task_version": "1"
        },
        "procgen-climber": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            12.6
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-climber",
          "task_version": "1"
        },
        "procgen-coinrun": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-coinrun",
          "task_version": "1"
        },
        "procgen-dodgeball": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            19
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-dodgeball",
          "task_version": "1"
        },
        "procgen-fruitbot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -1.5,
            32.4
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-fruitbot",
          "task_version": "1"
        },
        "procgen-heist": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-heist",
          "task_version": "1"
        },
        "procgen-jumper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-jumper",
          "task_version": "1"
        },
        "procgen-leaper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-leaper",
          "task_version": "1"
        },
        "procgen-maze": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-maze",
          "task_version": "1"
        },
        "procgen-miner": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-miner",
          "task_version": "1"
        },
        "procgen-ninja": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-ninja",
          "task_version": "1"
        },
        "procgen-plunder": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            4.5,
            30
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-plunder",
          "task_version": "1"
        },
        "procgen-starpilot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2.5,
            64
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-starpilot",
          "task_version": "1"
        },
        "crafter": {
          "engine": "crafter",
          "engine_version": "1.8.3",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            22
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "crafter",
          "task_version": "2"
        },
        "miniworld-oneroom": {
          "engine": "miniworld",
          "engine_version": "2.1.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "miniworld-oneroom",
          "task_version": "1"
        },
        "pettingzoo-pistonball": {
          "engine": "pettingzoo",
          "engine_version": "1.27.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -100,
            100
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "pettingzoo-pistonball",
          "task_version": "1"
        },
        "atari-breakout": {
          "engine": "ale-py",
          "engine_version": "0.12.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            864
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "atari-breakout",
          "task_version": "1"
        },
        "retro-airstriker": {
          "engine": "stable-retro",
          "engine_version": "1.0.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            10000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "retro-airstriker",
          "task_version": "1"
        },
        "nethack-score": {
          "engine": "nle",
          "engine_version": "1.3.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "nethack-score",
          "task_version": "1"
        },
        "minihack-room": {
          "engine": "minihack",
          "engine_version": "1.0.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "minihack-room",
          "task_version": "1"
        },
        "unity-food-collector": {
          "assets_sha256": "c28e27b2d3af4391a3e1e95b0871e01a16ef657ee062bb91e7740ac9d0c99be5",
          "camera_size": [
            84,
            84
          ],
          "capture_frame_rate": 30,
          "controlled_agent": "lowest-initial-agent-id",
          "engine": "mlagents-envs",
          "engine_version": "1.1.0",
          "numpy_version": "1.23.5",
          "observation": "rendered-rgb-only",
          "peer_policy": "no-op",
          "replay": "exact-frame-and-score",
          "reward_definition": "sum of native food rewards; green +1, red -1",
          "scene": "VisualFoodCollector",
          "score_range": [
            0,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "unity-food-collector",
          "task_version": "1"
        },
        "football-empty-goal": {
          "camera_size": [
            320,
            180
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:8774faa312fa0791f01966ef3152d23c012dda5117fd01366972a0bebfc13635",
          "control_policy": "release movement and shot before each decision",
          "engine": "gfootball",
          "engine_revision": "ba9952130898bde2879b01ff2b49d3d48d26ce81",
          "engine_version": "2.10.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native goal reward, no checkpoint shaping",
          "scenario": "academy_empty_goal_close",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "football-empty-goal",
          "task_version": "1"
        },
        "supertuxkart-lighthouse": {
          "assets_sha256": "56cd754653224efde4a40aef0400a71f03f72f18d2c35a89896db052dc096b23",
          "camera_size": [
            320,
            240
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:257166494f158e6b5b3441d4b52f99b83e8cef4ff60c6ce5b333251aa7cf4738",
          "engine": "PySuperTuxKart2",
          "engine_revision": "dd70f6823f248ae1df2ce513839a9b2c8c940c39",
          "engine_variant": "deterministic-render-v2",
          "engine_version": "0.7.4+ggbrender2",
          "neutral_preroll_steps": 50,
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "Mesa software / deterministic-render-v2",
          "renderer_threads": 1,
          "renderer_vector_bits": 128,
          "replay": "exact-frame-and-score",
          "reward_definition": "max(0, native overall distance) / native track length",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_seconds": 0.2,
          "task_id": "supertuxkart-lighthouse",
          "task_version": "3",
          "track": "lighthouse"
        },
        "luanti-chop-tree": {
          "camera_size": [
            64,
            64
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:f3f2df899b8709a904350472825f3253fd1a46a8d913322634a604e389fbb0fe",
          "engine": "craftium",
          "engine_revision": "8cffe4176e793f78d00b17fa7e46ccf333a7b5b0",
          "engine_variant": "serial-lockstep-v2",
          "engine_version": "0.0.1",
          "mesh_generation_threads": 1,
          "numpy_version": "2.5.3",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native Lua tree-node dig reward; first reward completes task",
          "scenario": "Craftium/ChopTree-v0: first tree block",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "pinned world snapshot; fixed map/Lua seed on initial reset",
          "simulation": "lockstep",
          "step_seconds": 0.02,
          "task_id": "luanti-chop-tree",
          "task_version": "1",
          "weather": "disabled via upstream enable_weather setting"
        },
        "supertux-first-coin": {
          "camera_size": [
            640,
            480
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:b3f936e8da2a38c0eab38ea1acfe56c1c47829cbe1f288775a135f0a67faadbf",
          "engine": "SuperTux",
          "engine_revision": "c1ddb4f28c54de77f07aa9965231a91c0ed06708",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.6.3+ggbench1",
          "frames_per_action": 8,
          "initialization": "skip level-introduction menu and fades; fresh player/save",
          "level": "world1/welcome_antarctica.stl",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "physics_step_seconds": 0.015,
          "renderer": "SDL software",
          "replay": "exact-frame-and-score",
          "reward_definition": "native level coin count; first coin completes task",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "supertux-first-coin",
          "task_version": "1"
        },
        "dcss-first-experience": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:942c43c70e0682b3d612e7c0b6e42eb8b35aeef42aad8bb77982bbbad4d617b8",
          "control_policy": "one native key per decision; no auto-explore or auto-fight",
          "engine": "DCSS",
          "engine_revision": "19cb0921cbcb6463fcedd1bb06636446c50b1905",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.34.0+ggbench1",
          "initialization": "advance title and seed-confirmation menus before observing",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 tiles / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native experience gain; death ends task",
          "scenario": "Dungeon:1; Minotaur Fighter with war axe; first experience",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "native uint64 seed; benchmark zero maps to uint64 maximum",
          "simulation": "lockstep",
          "task_id": "dcss-first-experience",
          "task_version": "1"
        },
        "openttd-first-road": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:227f7a13d0f6a96f2574ac0b428755101ad54def2217d629d08d059f4e2996b0",
          "cursor_step_pixels": 32,
          "engine": "OpenTTD",
          "engine_revision": "14ec60f248547d4d062a1160f0fc26d742319888",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "15.3+ggbench1",
          "initialization": "16 native ticks; fresh company and configuration",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native 32bpp software framebuffer; OpenGFX 7.1",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when native company infrastructure owns at least two road bits",
          "scenario": "new 64x64 company in 1950; first owned road",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 30,
          "task_id": "openttd-first-road",
          "task_version": "1"
        },
        "mindustry-copper": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:90468c4e5dde3dfaf18e1ec5fe337b4d32906271d3a7be899efd5b87756082a2",
          "cursor_step_pixels": 32,
          "engine": "Mindustry",
          "engine_revision": "067c720a8817c1c9fb586c03898a7d948caaed56",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "160.5+ggbench1",
          "frames_per_action": 6,
          "initialization": "120 native frames; fresh player and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "optional_effects": false,
          "physics_step_seconds": 0.016666666666666666,
          "renderer": "native SDL/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "native player-plus-core copper gain from initial total; target 15",
          "scenario": "serpulo/groundZero; native first-15-copper tutorial objective",
          "score_range": [
            0,
            15
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "mindustry-copper",
          "task_version": "1"
        },
        "cdda-first-weapon": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:eee69b35296e92b6ddc84a9ffbf737ab66ab48ae42dbbc3f1ee8e236b97c1b1f",
          "control_policy": "one native key per decision; tutorial popups remain visible",
          "engine": "Cataclysm: Dark Days Ahead",
          "engine_revision": "7b2efa5cea38e4d4d97dd0e63b28b9148623da59",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.I-1+ggbench1",
          "initialization": "native tutorial world and preset character; fresh user directory",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 software / UltimateCataclysm tiles",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when the native avatar wields the tutorial baseball bat",
          "scenario": "native tutorial; first wielded baseball bat",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "cdda-first-weapon",
          "task_version": "1"
        },
        "warzone-first-derrick": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:c2c0aee72affd4e5d5be6aa03659c533e3c968365e0f62eccf1f9fbdd30f6c22",
          "cursor_step_pixels": 32,
          "engine": "Warzone 2100",
          "engine_revision": "7d3d8e51a61ceee89e6bf672a41337e1a9cd7392",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "4.7.0+ggbench1",
          "fine_cursor_step_pixels": 8,
          "initialization": "40 native frames; fresh tutorial and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL3/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native completed structure; two unit losses end task",
          "scenario": "native TUTORIAL3; first constructed oil derrick",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 100,
          "task_id": "warzone-first-derrick",
          "task_version": "1"
        }
      },
      "decisions": 337,
      "p50_ms": 175.731291,
      "p95_ms": 663.8795995999999,
      "max_ms": 817.815375,
      "misses": 109,
      "latency_eligible": false,
      "latency_policy": {
        "id": "suite-p95-200-v1",
        "metric": "suite_p95_ms",
        "limit_ms": 200.0,
        "comparison": "strictly_less_than",
        "quantile_method": "linear",
        "sample_scope": "all_recorded_decisions_in_complete_suite"
      },
      "response_timeout_seconds": 120.0,
      "trust": "local-unattested",
      "official_rank": null,
      "per_game": {
        "coin-run": 0.0,
        "dodge-lanes": 100.0,
        "doom-basic": 68.571,
        "doom-defend-center": 4.762,
        "doom-defend-line": 9.524,
        "doom-take-cover": 100.0,
        "doom-health": 100.0,
        "doom-corridor": 9.091,
        "doom-home": 0.0,
        "doom-predict": 76.183,
        "procgen-bigfish": 0.0,
        "procgen-bossfight": 0.0,
        "procgen-caveflyer": 0.0,
        "procgen-chaser": 0.0,
        "procgen-climber": 0.0,
        "procgen-coinrun": 0.0,
        "procgen-dodgeball": 0.0,
        "procgen-fruitbot": 7.375,
        "procgen-heist": 0.0,
        "procgen-jumper": 0.0,
        "procgen-leaper": 0.0,
        "procgen-maze": 0.0,
        "procgen-miner": 0.0,
        "procgen-ninja": 0.0,
        "procgen-plunder": 0.0,
        "procgen-starpilot": 0.0,
        "crafter": 0.0,
        "miniworld-oneroom": 0.0,
        "pettingzoo-pistonball": 69.988,
        "atari-breakout": 0.0,
        "retro-airstriker": 0.0,
        "nethack-score": 0.0,
        "minihack-room": 0.0,
        "unity-food-collector": 0.0,
        "football-empty-goal": 0.0,
        "supertuxkart-lighthouse": 1.362,
        "luanti-chop-tree": 0.0,
        "supertux-first-coin": 0.0,
        "dcss-first-experience": 0.0,
        "openttd-first-road": 0.0,
        "mindustry-copper": 0.0,
        "cdda-first-weapon": 0.0,
        "warzone-first-derrick": 0.0
      },
      "errors": 1,
      "aborted_episodes": 1,
      "model": "LiquidAI/LFM2.5-VL-450M-MLX-4bit",
      "provider_metadata": {
        "transport": "persistent-mlx-jsonl",
        "requested_model": "LiquidAI/LFM2.5-VL-450M-MLX-4bit",
        "revision": "f19926f17a25164d4cbcdc16d9eaf4714b807cfb",
        "observation": "png-only",
        "temperature": 0.0,
        "thinking": false,
        "tool_events": 0,
        "prompt_version": "pixels-json-action/2",
        "warmup": "synthetic-gray-3/1"
      },
      "hardware_details": "Apple M3 Max; 48 GiB RAM; 16 CPU cores; macOS arm64. One local MLX model and one game run at a time. Weights remain loaded between games. Each model receives three gray startup images before scored actions. Startup image sizes: 160x120, 320x240, 960x640. Startup has no game data. Every game action remains in the measured result. Normal desktop tasks can use the same Mac."
    },
    {
      "agent": "mlx-mlx-community-SmolVLM2-256M-Video-Instruct-mlx",
      "mode": "exhibition",
      "hardware": "Darwin arm64",
      "version": "0.5.0",
      "max_steps": 8,
      "score": 0.211,
      "ci95": null,
      "episodes": 43,
      "seeds": 1,
      "seed_ids": [
        3000
      ],
      "games": [
        "coin-run",
        "dodge-lanes",
        "doom-basic",
        "doom-defend-center",
        "doom-defend-line",
        "doom-take-cover",
        "doom-health",
        "doom-corridor",
        "doom-home",
        "doom-predict",
        "procgen-bigfish",
        "procgen-bossfight",
        "procgen-caveflyer",
        "procgen-chaser",
        "procgen-climber",
        "procgen-coinrun",
        "procgen-dodgeball",
        "procgen-fruitbot",
        "procgen-heist",
        "procgen-jumper",
        "procgen-leaper",
        "procgen-maze",
        "procgen-miner",
        "procgen-ninja",
        "procgen-plunder",
        "procgen-starpilot",
        "crafter",
        "miniworld-oneroom",
        "pettingzoo-pistonball",
        "atari-breakout",
        "retro-airstriker",
        "nethack-score",
        "minihack-room",
        "unity-food-collector",
        "football-empty-goal",
        "supertuxkart-lighthouse",
        "luanti-chop-tree",
        "supertux-first-coin",
        "dcss-first-experience",
        "openttd-first-road",
        "mindustry-copper",
        "cdda-first-weapon",
        "warzone-first-derrick"
      ],
      "task_metadata": {
        "coin-run": {},
        "dodge-lanes": {},
        "doom-basic": {},
        "doom-defend-center": {},
        "doom-defend-line": {},
        "doom-take-cover": {},
        "doom-health": {},
        "doom-corridor": {},
        "doom-home": {},
        "doom-predict": {},
        "procgen-bigfish": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            40
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bigfish",
          "task_version": "1"
        },
        "procgen-bossfight": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bossfight",
          "task_version": "1"
        },
        "procgen-caveflyer": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2,
            12
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-caveflyer",
          "task_version": "1"
        },
        "procgen-chaser": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-chaser",
          "task_version": "1"
        },
        "procgen-climber": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            12.6
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-climber",
          "task_version": "1"
        },
        "procgen-coinrun": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-coinrun",
          "task_version": "1"
        },
        "procgen-dodgeball": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            19
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-dodgeball",
          "task_version": "1"
        },
        "procgen-fruitbot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -1.5,
            32.4
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-fruitbot",
          "task_version": "1"
        },
        "procgen-heist": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-heist",
          "task_version": "1"
        },
        "procgen-jumper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-jumper",
          "task_version": "1"
        },
        "procgen-leaper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-leaper",
          "task_version": "1"
        },
        "procgen-maze": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-maze",
          "task_version": "1"
        },
        "procgen-miner": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-miner",
          "task_version": "1"
        },
        "procgen-ninja": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-ninja",
          "task_version": "1"
        },
        "procgen-plunder": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            4.5,
            30
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-plunder",
          "task_version": "1"
        },
        "procgen-starpilot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2.5,
            64
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-starpilot",
          "task_version": "1"
        },
        "crafter": {
          "engine": "crafter",
          "engine_version": "1.8.3",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            22
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "crafter",
          "task_version": "2"
        },
        "miniworld-oneroom": {
          "engine": "miniworld",
          "engine_version": "2.1.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "miniworld-oneroom",
          "task_version": "1"
        },
        "pettingzoo-pistonball": {
          "engine": "pettingzoo",
          "engine_version": "1.27.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -100,
            100
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "pettingzoo-pistonball",
          "task_version": "1"
        },
        "atari-breakout": {
          "engine": "ale-py",
          "engine_version": "0.12.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            864
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "atari-breakout",
          "task_version": "1"
        },
        "retro-airstriker": {
          "engine": "stable-retro",
          "engine_version": "1.0.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            10000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "retro-airstriker",
          "task_version": "1"
        },
        "nethack-score": {
          "engine": "nle",
          "engine_version": "1.3.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "nethack-score",
          "task_version": "1"
        },
        "minihack-room": {
          "engine": "minihack",
          "engine_version": "1.0.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "minihack-room",
          "task_version": "1"
        },
        "unity-food-collector": {
          "assets_sha256": "c28e27b2d3af4391a3e1e95b0871e01a16ef657ee062bb91e7740ac9d0c99be5",
          "camera_size": [
            84,
            84
          ],
          "capture_frame_rate": 30,
          "controlled_agent": "lowest-initial-agent-id",
          "engine": "mlagents-envs",
          "engine_version": "1.1.0",
          "numpy_version": "1.23.5",
          "observation": "rendered-rgb-only",
          "peer_policy": "no-op",
          "replay": "exact-frame-and-score",
          "reward_definition": "sum of native food rewards; green +1, red -1",
          "scene": "VisualFoodCollector",
          "score_range": [
            0,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "unity-food-collector",
          "task_version": "1"
        },
        "football-empty-goal": {
          "camera_size": [
            320,
            180
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:8774faa312fa0791f01966ef3152d23c012dda5117fd01366972a0bebfc13635",
          "control_policy": "release movement and shot before each decision",
          "engine": "gfootball",
          "engine_revision": "ba9952130898bde2879b01ff2b49d3d48d26ce81",
          "engine_version": "2.10.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native goal reward, no checkpoint shaping",
          "scenario": "academy_empty_goal_close",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "football-empty-goal",
          "task_version": "1"
        },
        "supertuxkart-lighthouse": {
          "assets_sha256": "56cd754653224efde4a40aef0400a71f03f72f18d2c35a89896db052dc096b23",
          "camera_size": [
            320,
            240
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:257166494f158e6b5b3441d4b52f99b83e8cef4ff60c6ce5b333251aa7cf4738",
          "engine": "PySuperTuxKart2",
          "engine_revision": "dd70f6823f248ae1df2ce513839a9b2c8c940c39",
          "engine_variant": "deterministic-render-v2",
          "engine_version": "0.7.4+ggbrender2",
          "neutral_preroll_steps": 50,
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "Mesa software / deterministic-render-v2",
          "renderer_threads": 1,
          "renderer_vector_bits": 128,
          "replay": "exact-frame-and-score",
          "reward_definition": "max(0, native overall distance) / native track length",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_seconds": 0.2,
          "task_id": "supertuxkart-lighthouse",
          "task_version": "3",
          "track": "lighthouse"
        },
        "luanti-chop-tree": {
          "camera_size": [
            64,
            64
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:f3f2df899b8709a904350472825f3253fd1a46a8d913322634a604e389fbb0fe",
          "engine": "craftium",
          "engine_revision": "8cffe4176e793f78d00b17fa7e46ccf333a7b5b0",
          "engine_variant": "serial-lockstep-v2",
          "engine_version": "0.0.1",
          "mesh_generation_threads": 1,
          "numpy_version": "2.5.3",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native Lua tree-node dig reward; first reward completes task",
          "scenario": "Craftium/ChopTree-v0: first tree block",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "pinned world snapshot; fixed map/Lua seed on initial reset",
          "simulation": "lockstep",
          "step_seconds": 0.02,
          "task_id": "luanti-chop-tree",
          "task_version": "1",
          "weather": "disabled via upstream enable_weather setting"
        },
        "supertux-first-coin": {
          "camera_size": [
            640,
            480
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:b3f936e8da2a38c0eab38ea1acfe56c1c47829cbe1f288775a135f0a67faadbf",
          "engine": "SuperTux",
          "engine_revision": "c1ddb4f28c54de77f07aa9965231a91c0ed06708",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.6.3+ggbench1",
          "frames_per_action": 8,
          "initialization": "skip level-introduction menu and fades; fresh player/save",
          "level": "world1/welcome_antarctica.stl",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "physics_step_seconds": 0.015,
          "renderer": "SDL software",
          "replay": "exact-frame-and-score",
          "reward_definition": "native level coin count; first coin completes task",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "supertux-first-coin",
          "task_version": "1"
        },
        "dcss-first-experience": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:942c43c70e0682b3d612e7c0b6e42eb8b35aeef42aad8bb77982bbbad4d617b8",
          "control_policy": "one native key per decision; no auto-explore or auto-fight",
          "engine": "DCSS",
          "engine_revision": "19cb0921cbcb6463fcedd1bb06636446c50b1905",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.34.0+ggbench1",
          "initialization": "advance title and seed-confirmation menus before observing",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 tiles / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native experience gain; death ends task",
          "scenario": "Dungeon:1; Minotaur Fighter with war axe; first experience",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "native uint64 seed; benchmark zero maps to uint64 maximum",
          "simulation": "lockstep",
          "task_id": "dcss-first-experience",
          "task_version": "1"
        },
        "openttd-first-road": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:227f7a13d0f6a96f2574ac0b428755101ad54def2217d629d08d059f4e2996b0",
          "cursor_step_pixels": 32,
          "engine": "OpenTTD",
          "engine_revision": "14ec60f248547d4d062a1160f0fc26d742319888",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "15.3+ggbench1",
          "initialization": "16 native ticks; fresh company and configuration",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native 32bpp software framebuffer; OpenGFX 7.1",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when native company infrastructure owns at least two road bits",
          "scenario": "new 64x64 company in 1950; first owned road",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 30,
          "task_id": "openttd-first-road",
          "task_version": "1"
        },
        "mindustry-copper": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:90468c4e5dde3dfaf18e1ec5fe337b4d32906271d3a7be899efd5b87756082a2",
          "cursor_step_pixels": 32,
          "engine": "Mindustry",
          "engine_revision": "067c720a8817c1c9fb586c03898a7d948caaed56",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "160.5+ggbench1",
          "frames_per_action": 6,
          "initialization": "120 native frames; fresh player and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "optional_effects": false,
          "physics_step_seconds": 0.016666666666666666,
          "renderer": "native SDL/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "native player-plus-core copper gain from initial total; target 15",
          "scenario": "serpulo/groundZero; native first-15-copper tutorial objective",
          "score_range": [
            0,
            15
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "mindustry-copper",
          "task_version": "1"
        },
        "cdda-first-weapon": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:eee69b35296e92b6ddc84a9ffbf737ab66ab48ae42dbbc3f1ee8e236b97c1b1f",
          "control_policy": "one native key per decision; tutorial popups remain visible",
          "engine": "Cataclysm: Dark Days Ahead",
          "engine_revision": "7b2efa5cea38e4d4d97dd0e63b28b9148623da59",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.I-1+ggbench1",
          "initialization": "native tutorial world and preset character; fresh user directory",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 software / UltimateCataclysm tiles",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when the native avatar wields the tutorial baseball bat",
          "scenario": "native tutorial; first wielded baseball bat",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "cdda-first-weapon",
          "task_version": "1"
        },
        "warzone-first-derrick": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:c2c0aee72affd4e5d5be6aa03659c533e3c968365e0f62eccf1f9fbdd30f6c22",
          "cursor_step_pixels": 32,
          "engine": "Warzone 2100",
          "engine_revision": "7d3d8e51a61ceee89e6bf672a41337e1a9cd7392",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "4.7.0+ggbench1",
          "fine_cursor_step_pixels": 8,
          "initialization": "40 native frames; fresh tutorial and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL3/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native completed structure; two unit losses end task",
          "scenario": "native TUTORIAL3; first constructed oil derrick",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 100,
          "task_id": "warzone-first-derrick",
          "task_version": "1"
        }
      },
      "decisions": 57,
      "p50_ms": 1163.184958,
      "p95_ms": 1565.5551666,
      "max_ms": 1940.889875,
      "misses": 57,
      "latency_eligible": false,
      "latency_policy": {
        "id": "suite-p95-200-v1",
        "metric": "suite_p95_ms",
        "limit_ms": 200.0,
        "comparison": "strictly_less_than",
        "quantile_method": "linear",
        "sample_scope": "all_recorded_decisions_in_complete_suite"
      },
      "response_timeout_seconds": 120.0,
      "trust": "local-unattested",
      "official_rank": null,
      "per_game": {
        "coin-run": 0.0,
        "dodge-lanes": 0.0,
        "doom-basic": 0.0,
        "doom-defend-center": 0.0,
        "doom-defend-line": 0.0,
        "doom-take-cover": 0.0,
        "doom-health": 0.0,
        "doom-corridor": 9.091,
        "doom-home": 0.0,
        "doom-predict": 0.0,
        "procgen-bigfish": 0.0,
        "procgen-bossfight": 0.0,
        "procgen-caveflyer": 0.0,
        "procgen-chaser": 0.0,
        "procgen-climber": 0.0,
        "procgen-coinrun": 0.0,
        "procgen-dodgeball": 0.0,
        "procgen-fruitbot": 0.0,
        "procgen-heist": 0.0,
        "procgen-jumper": 0.0,
        "procgen-leaper": 0.0,
        "procgen-maze": 0.0,
        "procgen-miner": 0.0,
        "procgen-ninja": 0.0,
        "procgen-plunder": 0.0,
        "procgen-starpilot": 0.0,
        "crafter": 0.0,
        "miniworld-oneroom": 0.0,
        "pettingzoo-pistonball": 0.0,
        "atari-breakout": 0.0,
        "retro-airstriker": 0.0,
        "nethack-score": 0.0,
        "minihack-room": 0.0,
        "unity-food-collector": 0.0,
        "football-empty-goal": 0.0,
        "supertuxkart-lighthouse": 0.0,
        "luanti-chop-tree": 0.0,
        "supertux-first-coin": 0.0,
        "dcss-first-experience": 0.0,
        "openttd-first-road": 0.0,
        "mindustry-copper": 0.0,
        "cdda-first-weapon": 0.0,
        "warzone-first-derrick": 0.0
      },
      "errors": 41,
      "aborted_episodes": 41,
      "model": "mlx-community/SmolVLM2-256M-Video-Instruct-mlx",
      "provider_metadata": {
        "transport": "persistent-mlx-jsonl",
        "requested_model": "mlx-community/SmolVLM2-256M-Video-Instruct-mlx",
        "revision": "79901655c1a3d7ed6646c91325e3fde47e40e611",
        "observation": "png-only",
        "temperature": 0.0,
        "thinking": false,
        "tool_events": 0,
        "prompt_version": "pixels-json-action/2",
        "warmup": "synthetic-gray-3/1"
      },
      "hardware_details": "Apple M3 Max; 48 GiB RAM; 16 CPU cores; macOS arm64. One local MLX model and one game run at a time. Weights remain loaded between games. Each model receives three gray startup images before scored actions. Startup image sizes: 160x120, 320x240, 960x640. Startup has no game data. Every game action remains in the measured result. Normal desktop tasks can use the same Mac."
    },
    {
      "agent": "mlx-mlx-community-FastVLM-0-5B-bf16",
      "mode": "exhibition",
      "hardware": "Darwin arm64",
      "version": "0.5.0",
      "max_steps": 8,
      "score": 0.0,
      "ci95": null,
      "episodes": 43,
      "seeds": 1,
      "seed_ids": [
        3000
      ],
      "games": [
        "coin-run",
        "dodge-lanes",
        "doom-basic",
        "doom-defend-center",
        "doom-defend-line",
        "doom-take-cover",
        "doom-health",
        "doom-corridor",
        "doom-home",
        "doom-predict",
        "procgen-bigfish",
        "procgen-bossfight",
        "procgen-caveflyer",
        "procgen-chaser",
        "procgen-climber",
        "procgen-coinrun",
        "procgen-dodgeball",
        "procgen-fruitbot",
        "procgen-heist",
        "procgen-jumper",
        "procgen-leaper",
        "procgen-maze",
        "procgen-miner",
        "procgen-ninja",
        "procgen-plunder",
        "procgen-starpilot",
        "crafter",
        "miniworld-oneroom",
        "pettingzoo-pistonball",
        "atari-breakout",
        "retro-airstriker",
        "nethack-score",
        "minihack-room",
        "unity-food-collector",
        "football-empty-goal",
        "supertuxkart-lighthouse",
        "luanti-chop-tree",
        "supertux-first-coin",
        "dcss-first-experience",
        "openttd-first-road",
        "mindustry-copper",
        "cdda-first-weapon",
        "warzone-first-derrick"
      ],
      "task_metadata": {
        "coin-run": {},
        "dodge-lanes": {},
        "doom-basic": {},
        "doom-defend-center": {},
        "doom-defend-line": {},
        "doom-take-cover": {},
        "doom-health": {},
        "doom-corridor": {},
        "doom-home": {},
        "doom-predict": {},
        "procgen-bigfish": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            40
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bigfish",
          "task_version": "1"
        },
        "procgen-bossfight": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bossfight",
          "task_version": "1"
        },
        "procgen-caveflyer": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2,
            12
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-caveflyer",
          "task_version": "1"
        },
        "procgen-chaser": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-chaser",
          "task_version": "1"
        },
        "procgen-climber": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            12.6
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-climber",
          "task_version": "1"
        },
        "procgen-coinrun": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-coinrun",
          "task_version": "1"
        },
        "procgen-dodgeball": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            19
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-dodgeball",
          "task_version": "1"
        },
        "procgen-fruitbot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -1.5,
            32.4
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-fruitbot",
          "task_version": "1"
        },
        "procgen-heist": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-heist",
          "task_version": "1"
        },
        "procgen-jumper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-jumper",
          "task_version": "1"
        },
        "procgen-leaper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-leaper",
          "task_version": "1"
        },
        "procgen-maze": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-maze",
          "task_version": "1"
        },
        "procgen-miner": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-miner",
          "task_version": "1"
        },
        "procgen-ninja": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-ninja",
          "task_version": "1"
        },
        "procgen-plunder": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            4.5,
            30
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-plunder",
          "task_version": "1"
        },
        "procgen-starpilot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2.5,
            64
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-starpilot",
          "task_version": "1"
        },
        "crafter": {
          "engine": "crafter",
          "engine_version": "1.8.3",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            22
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "crafter",
          "task_version": "2"
        },
        "miniworld-oneroom": {
          "engine": "miniworld",
          "engine_version": "2.1.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "miniworld-oneroom",
          "task_version": "1"
        },
        "pettingzoo-pistonball": {
          "engine": "pettingzoo",
          "engine_version": "1.27.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -100,
            100
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "pettingzoo-pistonball",
          "task_version": "1"
        },
        "atari-breakout": {
          "engine": "ale-py",
          "engine_version": "0.12.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            864
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "atari-breakout",
          "task_version": "1"
        },
        "retro-airstriker": {
          "engine": "stable-retro",
          "engine_version": "1.0.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            10000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "retro-airstriker",
          "task_version": "1"
        },
        "nethack-score": {
          "engine": "nle",
          "engine_version": "1.3.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "nethack-score",
          "task_version": "1"
        },
        "minihack-room": {
          "engine": "minihack",
          "engine_version": "1.0.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "minihack-room",
          "task_version": "1"
        },
        "unity-food-collector": {
          "assets_sha256": "c28e27b2d3af4391a3e1e95b0871e01a16ef657ee062bb91e7740ac9d0c99be5",
          "camera_size": [
            84,
            84
          ],
          "capture_frame_rate": 30,
          "controlled_agent": "lowest-initial-agent-id",
          "engine": "mlagents-envs",
          "engine_version": "1.1.0",
          "numpy_version": "1.23.5",
          "observation": "rendered-rgb-only",
          "peer_policy": "no-op",
          "replay": "exact-frame-and-score",
          "reward_definition": "sum of native food rewards; green +1, red -1",
          "scene": "VisualFoodCollector",
          "score_range": [
            0,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "unity-food-collector",
          "task_version": "1"
        },
        "football-empty-goal": {
          "camera_size": [
            320,
            180
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:8774faa312fa0791f01966ef3152d23c012dda5117fd01366972a0bebfc13635",
          "control_policy": "release movement and shot before each decision",
          "engine": "gfootball",
          "engine_revision": "ba9952130898bde2879b01ff2b49d3d48d26ce81",
          "engine_version": "2.10.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native goal reward, no checkpoint shaping",
          "scenario": "academy_empty_goal_close",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "football-empty-goal",
          "task_version": "1"
        },
        "supertuxkart-lighthouse": {
          "assets_sha256": "56cd754653224efde4a40aef0400a71f03f72f18d2c35a89896db052dc096b23",
          "camera_size": [
            320,
            240
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:257166494f158e6b5b3441d4b52f99b83e8cef4ff60c6ce5b333251aa7cf4738",
          "engine": "PySuperTuxKart2",
          "engine_revision": "dd70f6823f248ae1df2ce513839a9b2c8c940c39",
          "engine_variant": "deterministic-render-v2",
          "engine_version": "0.7.4+ggbrender2",
          "neutral_preroll_steps": 50,
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "Mesa software / deterministic-render-v2",
          "renderer_threads": 1,
          "renderer_vector_bits": 128,
          "replay": "exact-frame-and-score",
          "reward_definition": "max(0, native overall distance) / native track length",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_seconds": 0.2,
          "task_id": "supertuxkart-lighthouse",
          "task_version": "3",
          "track": "lighthouse"
        },
        "luanti-chop-tree": {
          "camera_size": [
            64,
            64
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:f3f2df899b8709a904350472825f3253fd1a46a8d913322634a604e389fbb0fe",
          "engine": "craftium",
          "engine_revision": "8cffe4176e793f78d00b17fa7e46ccf333a7b5b0",
          "engine_variant": "serial-lockstep-v2",
          "engine_version": "0.0.1",
          "mesh_generation_threads": 1,
          "numpy_version": "2.5.3",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native Lua tree-node dig reward; first reward completes task",
          "scenario": "Craftium/ChopTree-v0: first tree block",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "pinned world snapshot; fixed map/Lua seed on initial reset",
          "simulation": "lockstep",
          "step_seconds": 0.02,
          "task_id": "luanti-chop-tree",
          "task_version": "1",
          "weather": "disabled via upstream enable_weather setting"
        },
        "supertux-first-coin": {
          "camera_size": [
            640,
            480
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:b3f936e8da2a38c0eab38ea1acfe56c1c47829cbe1f288775a135f0a67faadbf",
          "engine": "SuperTux",
          "engine_revision": "c1ddb4f28c54de77f07aa9965231a91c0ed06708",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.6.3+ggbench1",
          "frames_per_action": 8,
          "initialization": "skip level-introduction menu and fades; fresh player/save",
          "level": "world1/welcome_antarctica.stl",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "physics_step_seconds": 0.015,
          "renderer": "SDL software",
          "replay": "exact-frame-and-score",
          "reward_definition": "native level coin count; first coin completes task",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "supertux-first-coin",
          "task_version": "1"
        },
        "dcss-first-experience": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:942c43c70e0682b3d612e7c0b6e42eb8b35aeef42aad8bb77982bbbad4d617b8",
          "control_policy": "one native key per decision; no auto-explore or auto-fight",
          "engine": "DCSS",
          "engine_revision": "19cb0921cbcb6463fcedd1bb06636446c50b1905",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.34.0+ggbench1",
          "initialization": "advance title and seed-confirmation menus before observing",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 tiles / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native experience gain; death ends task",
          "scenario": "Dungeon:1; Minotaur Fighter with war axe; first experience",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "native uint64 seed; benchmark zero maps to uint64 maximum",
          "simulation": "lockstep",
          "task_id": "dcss-first-experience",
          "task_version": "1"
        },
        "openttd-first-road": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:227f7a13d0f6a96f2574ac0b428755101ad54def2217d629d08d059f4e2996b0",
          "cursor_step_pixels": 32,
          "engine": "OpenTTD",
          "engine_revision": "14ec60f248547d4d062a1160f0fc26d742319888",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "15.3+ggbench1",
          "initialization": "16 native ticks; fresh company and configuration",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native 32bpp software framebuffer; OpenGFX 7.1",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when native company infrastructure owns at least two road bits",
          "scenario": "new 64x64 company in 1950; first owned road",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 30,
          "task_id": "openttd-first-road",
          "task_version": "1"
        },
        "mindustry-copper": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:90468c4e5dde3dfaf18e1ec5fe337b4d32906271d3a7be899efd5b87756082a2",
          "cursor_step_pixels": 32,
          "engine": "Mindustry",
          "engine_revision": "067c720a8817c1c9fb586c03898a7d948caaed56",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "160.5+ggbench1",
          "frames_per_action": 6,
          "initialization": "120 native frames; fresh player and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "optional_effects": false,
          "physics_step_seconds": 0.016666666666666666,
          "renderer": "native SDL/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "native player-plus-core copper gain from initial total; target 15",
          "scenario": "serpulo/groundZero; native first-15-copper tutorial objective",
          "score_range": [
            0,
            15
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "mindustry-copper",
          "task_version": "1"
        },
        "cdda-first-weapon": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:eee69b35296e92b6ddc84a9ffbf737ab66ab48ae42dbbc3f1ee8e236b97c1b1f",
          "control_policy": "one native key per decision; tutorial popups remain visible",
          "engine": "Cataclysm: Dark Days Ahead",
          "engine_revision": "7b2efa5cea38e4d4d97dd0e63b28b9148623da59",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.I-1+ggbench1",
          "initialization": "native tutorial world and preset character; fresh user directory",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 software / UltimateCataclysm tiles",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when the native avatar wields the tutorial baseball bat",
          "scenario": "native tutorial; first wielded baseball bat",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "cdda-first-weapon",
          "task_version": "1"
        },
        "warzone-first-derrick": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:c2c0aee72affd4e5d5be6aa03659c533e3c968365e0f62eccf1f9fbdd30f6c22",
          "cursor_step_pixels": 32,
          "engine": "Warzone 2100",
          "engine_revision": "7d3d8e51a61ceee89e6bf672a41337e1a9cd7392",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "4.7.0+ggbench1",
          "fine_cursor_step_pixels": 8,
          "initialization": "40 native frames; fresh tutorial and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL3/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native completed structure; two unit losses end task",
          "scenario": "native TUTORIAL3; first constructed oil derrick",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 100,
          "task_id": "warzone-first-derrick",
          "task_version": "1"
        }
      },
      "decisions": 50,
      "p50_ms": 679.0850625,
      "p95_ms": 1429.3062501499992,
      "max_ms": 1778.920875,
      "misses": 50,
      "latency_eligible": false,
      "latency_policy": {
        "id": "suite-p95-200-v1",
        "metric": "suite_p95_ms",
        "limit_ms": 200.0,
        "comparison": "strictly_less_than",
        "quantile_method": "linear",
        "sample_scope": "all_recorded_decisions_in_complete_suite"
      },
      "response_timeout_seconds": 120.0,
      "trust": "local-unattested",
      "official_rank": null,
      "per_game": {
        "coin-run": 0.0,
        "dodge-lanes": 0.0,
        "doom-basic": 0.0,
        "doom-defend-center": 0.0,
        "doom-defend-line": 0.0,
        "doom-take-cover": 0.0,
        "doom-health": 0.0,
        "doom-corridor": 0.0,
        "doom-home": 0.0,
        "doom-predict": 0.0,
        "procgen-bigfish": 0.0,
        "procgen-bossfight": 0.0,
        "procgen-caveflyer": 0.0,
        "procgen-chaser": 0.0,
        "procgen-climber": 0.0,
        "procgen-coinrun": 0.0,
        "procgen-dodgeball": 0.0,
        "procgen-fruitbot": 0.0,
        "procgen-heist": 0.0,
        "procgen-jumper": 0.0,
        "procgen-leaper": 0.0,
        "procgen-maze": 0.0,
        "procgen-miner": 0.0,
        "procgen-ninja": 0.0,
        "procgen-plunder": 0.0,
        "procgen-starpilot": 0.0,
        "crafter": 0.0,
        "miniworld-oneroom": 0.0,
        "pettingzoo-pistonball": 0.0,
        "atari-breakout": 0.0,
        "retro-airstriker": 0.0,
        "nethack-score": 0.0,
        "minihack-room": 0.0,
        "unity-food-collector": 0.0,
        "football-empty-goal": 0.0,
        "supertuxkart-lighthouse": 0.0,
        "luanti-chop-tree": 0.0,
        "supertux-first-coin": 0.0,
        "dcss-first-experience": 0.0,
        "openttd-first-road": 0.0,
        "mindustry-copper": 0.0,
        "cdda-first-weapon": 0.0,
        "warzone-first-derrick": 0.0
      },
      "errors": 42,
      "aborted_episodes": 42,
      "model": "mlx-community/FastVLM-0.5B-bf16",
      "provider_metadata": {
        "transport": "persistent-mlx-jsonl",
        "requested_model": "mlx-community/FastVLM-0.5B-bf16",
        "revision": "81ffe929046666c43de53691147b1669ba0f3a4c",
        "observation": "png-only",
        "temperature": 0.0,
        "thinking": false,
        "tool_events": 0,
        "prompt_version": "pixels-json-action/2",
        "warmup": "synthetic-gray-3/1"
      },
      "hardware_details": "Apple M3 Max; 48 GiB RAM; 16 CPU cores; macOS arm64. One local MLX model and one game run at a time. Weights remain loaded between games. Each model receives three gray startup images before scored actions. Startup image sizes: 160x120, 320x240, 960x640. Startup has no game data. Every game action remains in the measured result. Normal desktop tasks can use the same Mac."
    },
    {
      "agent": "mlx-mlx-community-SmolVLM2-500M-Video-Instruct-mlx",
      "mode": "exhibition",
      "hardware": "Darwin arm64",
      "version": "0.5.0",
      "max_steps": 8,
      "score": 0.0,
      "ci95": null,
      "episodes": 43,
      "seeds": 1,
      "seed_ids": [
        3000
      ],
      "games": [
        "coin-run",
        "dodge-lanes",
        "doom-basic",
        "doom-defend-center",
        "doom-defend-line",
        "doom-take-cover",
        "doom-health",
        "doom-corridor",
        "doom-home",
        "doom-predict",
        "procgen-bigfish",
        "procgen-bossfight",
        "procgen-caveflyer",
        "procgen-chaser",
        "procgen-climber",
        "procgen-coinrun",
        "procgen-dodgeball",
        "procgen-fruitbot",
        "procgen-heist",
        "procgen-jumper",
        "procgen-leaper",
        "procgen-maze",
        "procgen-miner",
        "procgen-ninja",
        "procgen-plunder",
        "procgen-starpilot",
        "crafter",
        "miniworld-oneroom",
        "pettingzoo-pistonball",
        "atari-breakout",
        "retro-airstriker",
        "nethack-score",
        "minihack-room",
        "unity-food-collector",
        "football-empty-goal",
        "supertuxkart-lighthouse",
        "luanti-chop-tree",
        "supertux-first-coin",
        "dcss-first-experience",
        "openttd-first-road",
        "mindustry-copper",
        "cdda-first-weapon",
        "warzone-first-derrick"
      ],
      "task_metadata": {
        "coin-run": {},
        "dodge-lanes": {},
        "doom-basic": {},
        "doom-defend-center": {},
        "doom-defend-line": {},
        "doom-take-cover": {},
        "doom-health": {},
        "doom-corridor": {},
        "doom-home": {},
        "doom-predict": {},
        "procgen-bigfish": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            40
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bigfish",
          "task_version": "1"
        },
        "procgen-bossfight": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-bossfight",
          "task_version": "1"
        },
        "procgen-caveflyer": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2,
            12
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-caveflyer",
          "task_version": "1"
        },
        "procgen-chaser": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-chaser",
          "task_version": "1"
        },
        "procgen-climber": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1,
            12.6
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-climber",
          "task_version": "1"
        },
        "procgen-coinrun": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-coinrun",
          "task_version": "1"
        },
        "procgen-dodgeball": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            19
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-dodgeball",
          "task_version": "1"
        },
        "procgen-fruitbot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -1.5,
            32.4
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-fruitbot",
          "task_version": "1"
        },
        "procgen-heist": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-heist",
          "task_version": "1"
        },
        "procgen-jumper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-jumper",
          "task_version": "1"
        },
        "procgen-leaper": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-leaper",
          "task_version": "1"
        },
        "procgen-maze": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-maze",
          "task_version": "1"
        },
        "procgen-miner": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            1.5,
            13
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-miner",
          "task_version": "1"
        },
        "procgen-ninja": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            3.5,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-ninja",
          "task_version": "1"
        },
        "procgen-plunder": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            4.5,
            30
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-plunder",
          "task_version": "1"
        },
        "procgen-starpilot": {
          "engine": "procgen",
          "engine_version": "0.10.7",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            2.5,
            64
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "procgen-starpilot",
          "task_version": "1"
        },
        "crafter": {
          "engine": "crafter",
          "engine_version": "1.8.3",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            22
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "crafter",
          "task_version": "2"
        },
        "miniworld-oneroom": {
          "engine": "miniworld",
          "engine_version": "2.1.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "miniworld-oneroom",
          "task_version": "1"
        },
        "pettingzoo-pistonball": {
          "engine": "pettingzoo",
          "engine_version": "1.27.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            -100,
            100
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "pettingzoo-pistonball",
          "task_version": "1"
        },
        "atari-breakout": {
          "engine": "ale-py",
          "engine_version": "0.12.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            864
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "atari-breakout",
          "task_version": "1"
        },
        "retro-airstriker": {
          "engine": "stable-retro",
          "engine_version": "1.0.1",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            10000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "retro-airstriker",
          "task_version": "1"
        },
        "nethack-score": {
          "engine": "nle",
          "engine_version": "1.3.0",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1000
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "nethack-score",
          "task_version": "1"
        },
        "minihack-room": {
          "engine": "minihack",
          "engine_version": "1.0.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "minihack-room",
          "task_version": "1"
        },
        "unity-food-collector": {
          "assets_sha256": "c28e27b2d3af4391a3e1e95b0871e01a16ef657ee062bb91e7740ac9d0c99be5",
          "camera_size": [
            84,
            84
          ],
          "capture_frame_rate": 30,
          "controlled_agent": "lowest-initial-agent-id",
          "engine": "mlagents-envs",
          "engine_version": "1.1.0",
          "numpy_version": "1.23.5",
          "observation": "rendered-rgb-only",
          "peer_policy": "no-op",
          "replay": "exact-frame-and-score",
          "reward_definition": "sum of native food rewards; green +1, red -1",
          "scene": "VisualFoodCollector",
          "score_range": [
            0,
            10
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "unity-food-collector",
          "task_version": "1"
        },
        "football-empty-goal": {
          "camera_size": [
            320,
            180
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:8774faa312fa0791f01966ef3152d23c012dda5117fd01366972a0bebfc13635",
          "control_policy": "release movement and shot before each decision",
          "engine": "gfootball",
          "engine_revision": "ba9952130898bde2879b01ff2b49d3d48d26ce81",
          "engine_version": "2.10.2",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native goal reward, no checkpoint shaping",
          "scenario": "academy_empty_goal_close",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "football-empty-goal",
          "task_version": "1"
        },
        "supertuxkart-lighthouse": {
          "assets_sha256": "56cd754653224efde4a40aef0400a71f03f72f18d2c35a89896db052dc096b23",
          "camera_size": [
            320,
            240
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:257166494f158e6b5b3441d4b52f99b83e8cef4ff60c6ce5b333251aa7cf4738",
          "engine": "PySuperTuxKart2",
          "engine_revision": "dd70f6823f248ae1df2ce513839a9b2c8c940c39",
          "engine_variant": "deterministic-render-v2",
          "engine_version": "0.7.4+ggbrender2",
          "neutral_preroll_steps": 50,
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "Mesa software / deterministic-render-v2",
          "renderer_threads": 1,
          "renderer_vector_bits": 128,
          "replay": "exact-frame-and-score",
          "reward_definition": "max(0, native overall distance) / native track length",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_seconds": 0.2,
          "task_id": "supertuxkart-lighthouse",
          "task_version": "3",
          "track": "lighthouse"
        },
        "luanti-chop-tree": {
          "camera_size": [
            64,
            64
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:f3f2df899b8709a904350472825f3253fd1a46a8d913322634a604e389fbb0fe",
          "engine": "craftium",
          "engine_revision": "8cffe4176e793f78d00b17fa7e46ccf333a7b5b0",
          "engine_variant": "serial-lockstep-v2",
          "engine_version": "0.0.1",
          "mesh_generation_threads": 1,
          "numpy_version": "2.5.3",
          "observation": "rendered-rgb-only",
          "replay": "exact-frame-and-score",
          "reward_definition": "native Lua tree-node dig reward; first reward completes task",
          "scenario": "Craftium/ChopTree-v0: first tree block",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "pinned world snapshot; fixed map/Lua seed on initial reset",
          "simulation": "lockstep",
          "step_seconds": 0.02,
          "task_id": "luanti-chop-tree",
          "task_version": "1",
          "weather": "disabled via upstream enable_weather setting"
        },
        "supertux-first-coin": {
          "camera_size": [
            640,
            480
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:b3f936e8da2a38c0eab38ea1acfe56c1c47829cbe1f288775a135f0a67faadbf",
          "engine": "SuperTux",
          "engine_revision": "c1ddb4f28c54de77f07aa9965231a91c0ed06708",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.6.3+ggbench1",
          "frames_per_action": 8,
          "initialization": "skip level-introduction menu and fades; fresh player/save",
          "level": "world1/welcome_antarctica.stl",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "physics_step_seconds": 0.015,
          "renderer": "SDL software",
          "replay": "exact-frame-and-score",
          "reward_definition": "native level coin count; first coin completes task",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "supertux-first-coin",
          "task_version": "1"
        },
        "dcss-first-experience": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:942c43c70e0682b3d612e7c0b6e42eb8b35aeef42aad8bb77982bbbad4d617b8",
          "control_policy": "one native key per decision; no auto-explore or auto-fight",
          "engine": "DCSS",
          "engine_revision": "19cb0921cbcb6463fcedd1bb06636446c50b1905",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.34.0+ggbench1",
          "initialization": "advance title and seed-confirmation menus before observing",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 tiles / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native experience gain; death ends task",
          "scenario": "Dungeon:1; Minotaur Fighter with war axe; first experience",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "native uint64 seed; benchmark zero maps to uint64 maximum",
          "simulation": "lockstep",
          "task_id": "dcss-first-experience",
          "task_version": "1"
        },
        "openttd-first-road": {
          "camera_size": [
            800,
            600
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:227f7a13d0f6a96f2574ac0b428755101ad54def2217d629d08d059f4e2996b0",
          "cursor_step_pixels": 32,
          "engine": "OpenTTD",
          "engine_revision": "14ec60f248547d4d062a1160f0fc26d742319888",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "15.3+ggbench1",
          "initialization": "16 native ticks; fresh company and configuration",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native 32bpp software framebuffer; OpenGFX 7.1",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when native company infrastructure owns at least two road bits",
          "scenario": "new 64x64 company in 1950; first owned road",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 30,
          "task_id": "openttd-first-road",
          "task_version": "1"
        },
        "mindustry-copper": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:90468c4e5dde3dfaf18e1ec5fe337b4d32906271d3a7be899efd5b87756082a2",
          "cursor_step_pixels": 32,
          "engine": "Mindustry",
          "engine_revision": "067c720a8817c1c9fb586c03898a7d948caaed56",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "160.5+ggbench1",
          "frames_per_action": 6,
          "initialization": "120 native frames; fresh player and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "optional_effects": false,
          "physics_step_seconds": 0.016666666666666666,
          "renderer": "native SDL/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "native player-plus-core copper gain from initial total; target 15",
          "scenario": "serpulo/groundZero; native first-15-copper tutorial objective",
          "score_range": [
            0,
            15
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "mindustry-copper",
          "task_version": "1"
        },
        "cdda-first-weapon": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:eee69b35296e92b6ddc84a9ffbf737ab66ab48ae42dbbc3f1ee8e236b97c1b1f",
          "control_policy": "one native key per decision; tutorial popups remain visible",
          "engine": "Cataclysm: Dark Days Ahead",
          "engine_revision": "7b2efa5cea38e4d4d97dd0e63b28b9148623da59",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "0.I-1+ggbench1",
          "initialization": "native tutorial world and preset character; fresh user directory",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL2 software / UltimateCataclysm tiles",
          "replay": "exact-frame-and-score",
          "reward_definition": "one when the native avatar wields the tutorial baseball bat",
          "scenario": "native tutorial; first wielded baseball bat",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "task_id": "cdda-first-weapon",
          "task_version": "1"
        },
        "warzone-first-derrick": {
          "camera_size": [
            960,
            640
          ],
          "container_architecture": "arm64",
          "container_image": "sha256:c2c0aee72affd4e5d5be6aa03659c533e3c968365e0f62eccf1f9fbdd30f6c22",
          "cursor_step_pixels": 32,
          "engine": "Warzone 2100",
          "engine_revision": "7d3d8e51a61ceee89e6bf672a41337e1a9cd7392",
          "engine_variant": "native-lockstep-v1",
          "engine_version": "4.7.0+ggbench1",
          "fine_cursor_step_pixels": 8,
          "initialization": "40 native frames; fresh tutorial and settings",
          "numpy_version": "1.26.4",
          "observation": "rendered-rgb-only",
          "renderer": "native SDL3/OpenGL / Xvfb / Mesa",
          "replay": "exact-frame-and-score",
          "reward_definition": "one on first native completed structure; two unit losses end task",
          "scenario": "native TUTORIAL3; first constructed oil derrick",
          "score_range": [
            0,
            1
          ],
          "seed_policy": "fixed-initial-reset",
          "simulation": "lockstep",
          "step_milliseconds": 100,
          "task_id": "warzone-first-derrick",
          "task_version": "1"
        }
      },
      "decisions": 43,
      "p50_ms": 1291.880167,
      "p95_ms": 1982.700783,
      "max_ms": 2168.158792,
      "misses": 43,
      "latency_eligible": false,
      "latency_policy": {
        "id": "suite-p95-200-v1",
        "metric": "suite_p95_ms",
        "limit_ms": 200.0,
        "comparison": "strictly_less_than",
        "quantile_method": "linear",
        "sample_scope": "all_recorded_decisions_in_complete_suite"
      },
      "response_timeout_seconds": 120.0,
      "trust": "local-unattested",
      "official_rank": null,
      "per_game": {
        "coin-run": 0.0,
        "dodge-lanes": 0.0,
        "doom-basic": 0.0,
        "doom-defend-center": 0.0,
        "doom-defend-line": 0.0,
        "doom-take-cover": 0.0,
        "doom-health": 0.0,
        "doom-corridor": 0.0,
        "doom-home": 0.0,
        "doom-predict": 0.0,
        "procgen-bigfish": 0.0,
        "procgen-bossfight": 0.0,
        "procgen-caveflyer": 0.0,
        "procgen-chaser": 0.0,
        "procgen-climber": 0.0,
        "procgen-coinrun": 0.0,
        "procgen-dodgeball": 0.0,
        "procgen-fruitbot": 0.0,
        "procgen-heist": 0.0,
        "procgen-jumper": 0.0,
        "procgen-leaper": 0.0,
        "procgen-maze": 0.0,
        "procgen-miner": 0.0,
        "procgen-ninja": 0.0,
        "procgen-plunder": 0.0,
        "procgen-starpilot": 0.0,
        "crafter": 0.0,
        "miniworld-oneroom": 0.0,
        "pettingzoo-pistonball": 0.0,
        "atari-breakout": 0.0,
        "retro-airstriker": 0.0,
        "nethack-score": 0.0,
        "minihack-room": 0.0,
        "unity-food-collector": 0.0,
        "football-empty-goal": 0.0,
        "supertuxkart-lighthouse": 0.0,
        "luanti-chop-tree": 0.0,
        "supertux-first-coin": 0.0,
        "dcss-first-experience": 0.0,
        "openttd-first-road": 0.0,
        "mindustry-copper": 0.0,
        "cdda-first-weapon": 0.0,
        "warzone-first-derrick": 0.0
      },
      "errors": 43,
      "aborted_episodes": 43,
      "model": "mlx-community/SmolVLM2-500M-Video-Instruct-mlx",
      "provider_metadata": {
        "transport": "persistent-mlx-jsonl",
        "requested_model": "mlx-community/SmolVLM2-500M-Video-Instruct-mlx",
        "revision": "fa57db46815177fbdfd65cc85a2b3416a8332268",
        "observation": "png-only",
        "temperature": 0.0,
        "thinking": false,
        "tool_events": 0,
        "prompt_version": "pixels-json-action/2",
        "warmup": "synthetic-gray-3/1"
      },
      "hardware_details": "Apple M3 Max; 48 GiB RAM; 16 CPU cores; macOS arm64. One local MLX model and one game run at a time. Weights remain loaded between games. Each model receives three gray startup images before scored actions. Startup image sizes: 160x120, 320x240, 960x640. Startup has no game data. Every game action remains in the measured result. Normal desktop tasks can use the same Mac."
    }
  ],
  "official": [],
  "catalog": [
    {
      "name": "Coin Run",
      "category": "ORIGINAL \u00b7 2D",
      "note": "Collect coins using only rendered pixels. Seeded 10\u00d710 world; fixed-horizon score.",
      "url": "https://github.com/DanielTea/generalgamebench",
      "status": "runnable",
      "id": "coin-run",
      "description": "Guide the cyan player around a grid and collect yellow coins. Each move tests visual localization and a short route to the next target.",
      "image": "media/coin-run.webp",
      "image_alt": "Coin Run: recorded coin-run gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "Daniel Tremer / GeneralGameBench",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded coin-run task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "video": "media/coin-run.mp4",
      "animation": "media/coin-run.gif",
      "integration_state": "validated",
      "task_ids": [
        "coin-run"
      ],
      "preview_task": "coin-run"
    },
    {
      "name": "Dodge Lanes",
      "category": "ORIGINAL \u00b7 2D",
      "note": "Avoid falling obstacles. Two local visual policies and controls are included.",
      "url": "https://github.com/DanielTea/generalgamebench",
      "status": "runnable",
      "id": "dodge-lanes",
      "description": "Dodge falling red blocks by moving between lanes. Staying alive requires noticing approaching obstacles and choosing a safe position.",
      "image": "media/dodge-lanes.webp",
      "image_alt": "Dodge Lanes: recorded dodge-lanes gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "Daniel Tremer / GeneralGameBench",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded dodge-lanes task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "video": "media/dodge-lanes.mp4",
      "animation": "media/dodge-lanes.gif",
      "integration_state": "validated",
      "task_ids": [
        "dodge-lanes"
      ],
      "preview_task": "dodge-lanes"
    },
    {
      "name": "ViZDoom \u00b7 eight scenarios",
      "category": "DOOM ENGINE \u00b7 3D",
      "note": "Basic, defend center, defend line, take cover, health gathering, corridor, home and prediction. These are eight scenarios of one game family.",
      "url": "https://github.com/Farama-Foundation/ViZDoom",
      "status": "runnable",
      "id": "vizdoom-eight-scenarios",
      "description": "First-person Doom tasks cover aiming, navigation, defense and survival. Eight scenarios share one game engine and present different control challenges.",
      "image": "media/vizdoom-eight-scenarios.webp",
      "image_alt": "ViZDoom \u00b7 eight scenarios: recorded doom-basic gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "GeneralGameBench capture; Doom content by id Software; ViZDoom contributors",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded doom-basic task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "video": "media/vizdoom-eight-scenarios.mp4",
      "animation": "media/vizdoom-eight-scenarios.gif",
      "integration_state": "validated",
      "task_ids": [
        "doom-basic",
        "doom-defend-center",
        "doom-defend-line",
        "doom-take-cover",
        "doom-health",
        "doom-corridor",
        "doom-home",
        "doom-predict"
      ],
      "preview_task": "doom-basic"
    },
    {
      "name": "SuperTuxKart",
      "category": "OPEN SOURCE \u00b7 3D RACING",
      "note": "The Lighthouse time trial uses task version 3 and renderer deterministic-render-v2. Tests passed on the Mac and native Linux x86_64. They compare exact images and scores over long action sequences. This task uses one kart and one track.",
      "url": "https://github.com/supertuxkart/stk-code",
      "status": "runnable",
      "id": "supertuxkart",
      "description": "Race karts through colorful tracks with drifting, obstacles and power-ups. The game also includes battle, soccer and collection modes.",
      "image": "media/supertuxkart.webp",
      "image_alt": "SuperTuxKart: recorded supertuxkart-lighthouse gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.3.0",
      "media_credit": "SuperTuxKart contributors",
      "media_kind": "Recorded benchmark",
      "media_caption": "This archived clip uses task version 2 and renderer deterministic-render-v1. The random reference policy controls the kart. Consecutive frames play at 6 fps. Playback speed does not measure reaction time.",
      "integration_state": "validated",
      "task_ids": [
        "supertuxkart-lighthouse"
      ],
      "video": "media/supertuxkart.mp4",
      "animation": "media/supertuxkart.gif",
      "preview_task": "supertuxkart-lighthouse"
    },
    {
      "name": "Luanti",
      "category": "OPEN SOURCE \u00b7 3D SANDBOX",
      "note": "Validated one-tree task in the pinned Craftium forest with an explicit serial-lockstep engine patch on Linux ARM64 and x86_64. Native tree reward and long exact camera/score replay pass; coordinates and voxels remain private. New task, outside the frozen v0.2 model standings.",
      "url": "https://github.com/luanti-org/luanti",
      "status": "runnable",
      "id": "luanti",
      "description": "Explore, dig and build in voxel worlds created with the Luanti engine. Its community games range from open-ended survival to focused platforming tasks.",
      "image": "media/luanti.webp",
      "image_alt": "Luanti: recorded luanti-chop-tree gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.3.0",
      "media_credit": "Luanti / Minetest Game contributors",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded luanti-chop-tree task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "integration_state": "validated",
      "task_ids": [
        "luanti-chop-tree"
      ],
      "video": "media/luanti.mp4",
      "animation": "media/luanti.gif",
      "preview_task": "luanti-chop-tree"
    },
    {
      "name": "Veloren",
      "category": "OPEN SOURCE \u00b7 3D RPG",
      "note": "Native Vulkan world, character creation, movement, interaction and camera work on this Mac. Exact delayed visual replay fails; no rewarded collection route is verified. The diagnostic is not ranked.",
      "url": "https://gitlab.com/veloren/veloren",
      "status": "candidate",
      "id": "veloren",
      "description": "Explore a voxel fantasy world, fight creatures and discover settlements. Movement, combat and long journeys create varied visual challenges.",
      "image": "media/veloren.webp",
      "image_alt": "Veloren: native diagnostic capture, unranked",
      "media_source": "https://github.com/DanielTea/generalgamebench/tree/v0.3.0/environments/experimental-veloren",
      "media_credit": "GeneralGameBench capture; Veloren contributors",
      "media_kind": "Native diagnostic \u00b7 unranked",
      "media_caption": "Native prototype capture. Exact visual replay failed; this candidate has no admitted task or ranking. Consecutive frames at 6 fps do not measure reaction time.",
      "integration_state": "experimental",
      "preview_url": "https://veloren.net/",
      "preview_label": "Watch project footage",
      "video": "media/veloren.mp4",
      "animation": "media/veloren.gif"
    },
    {
      "name": "Warzone 2100",
      "category": "OPEN SOURCE \u00b7 3D RTS",
      "note": "Build the first oil derrick in native TUTORIAL3 using truck selection and mouse controls. Native construction scoring and exact delayed replay are verified. This bounded task is outside the frozen v0.2 model standings.",
      "url": "https://github.com/Warzone2100/warzone2100",
      "status": "runnable",
      "id": "warzone-2100",
      "description": "Direct a customizable army in a post-apocalyptic real-time strategy game. Research, base building and positioning determine how battles unfold.",
      "image": "media/warzone-2100.webp",
      "image_alt": "Warzone 2100: recorded warzone-first-derrick gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.3.0",
      "media_credit": "GeneralGameBench capture; Pumpkin Studios, Underdone Gaming, Warzone 2100 Project",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded warzone-first-derrick task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "integration_state": "validated",
      "preview_url": "https://store.steampowered.com/app/1241950/",
      "preview_label": "Watch official trailer",
      "task_ids": [
        "warzone-first-derrick"
      ],
      "video": "media/warzone-2100.mp4",
      "animation": "media/warzone-2100.gif",
      "preview_task": "warzone-first-derrick"
    },
    {
      "name": "0 A.D.",
      "category": "OPEN SOURCE \u00b7 3D RTS",
      "note": "Native seeded Mainland map, standard mouse controls and wood-gathering scoring work. Display initialization and render clocks are patched, but exact visual replay still fails. The diagnostic is not ranked.",
      "url": "https://gitea.wildfiregames.com/0ad/0ad",
      "status": "candidate",
      "id": "0-a-d",
      "description": "Build an ancient civilization, gather resources and lead armies into battle. Economy and military decisions compete for attention in real time.",
      "image": "media/0-a-d.webp",
      "image_alt": "0 A.D.: native diagnostic capture, unranked",
      "media_source": "https://github.com/DanielTea/generalgamebench/tree/v0.3.0/environments/experimental-0ad",
      "media_credit": "GeneralGameBench capture; Wildfire Games / 0 A.D. contributors",
      "media_kind": "Native diagnostic \u00b7 unranked",
      "media_caption": "Native prototype capture. Exact visual replay failed; this candidate has no admitted task or ranking. Consecutive frames at 6 fps do not measure reaction time.",
      "integration_state": "experimental",
      "preview_url": "https://play0ad.com/media/screenshots/",
      "preview_label": "View game gallery",
      "video": "media/0-a-d.mp4",
      "animation": "media/0-a-d.gif"
    },
    {
      "name": "SuperTux",
      "category": "OPEN SOURCE \u00b7 2D PLATFORMER",
      "note": "Native Welcome to Antarctica level: collect the first coin with normal run/jump controls. Coin, death and delayed exact camera/score replay are verified on the pinned SDL build. This bounded task is outside the frozen v0.2 model standings.",
      "url": "https://github.com/SuperTux/supertux",
      "status": "runnable",
      "id": "supertux",
      "description": "Run and jump through side-scrolling levels as Tux the penguin. Avoid enemies and hazards while exploring paths and collecting items.",
      "image": "media/supertux.webp",
      "image_alt": "SuperTux: recorded supertux-first-coin gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.3.0",
      "media_credit": "GeneralGameBench capture; The SuperTux Team",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded supertux-first-coin task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "integration_state": "validated",
      "preview_url": "https://store.steampowered.com/app/1572920/",
      "preview_label": "Watch official trailer",
      "task_ids": [
        "supertux-first-coin"
      ],
      "video": "media/supertux.mp4",
      "animation": "media/supertux.gif",
      "preview_task": "supertux-first-coin"
    },
    {
      "name": "Mindustry",
      "category": "INDIE \u00b7 2D STRATEGY",
      "note": "Mine 15 copper in native Ground Zero with movement and cursor controls. The pinned SDL clients for Linux ARM64 and x86_64 pass native reward and delayed exact camera replay. This bounded task is outside the frozen v0.2 model standings.",
      "url": "https://github.com/Anuken/Mindustry",
      "status": "runnable",
      "id": "mindustry",
      "description": "Build supply chains that feed factories and defensive turrets. Balancing production, logistics and enemy waves is central to survival.",
      "image": "media/mindustry.webp",
      "image_alt": "Mindustry: recorded mindustry-copper gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.3.0",
      "media_credit": "GeneralGameBench capture; AnukenDev",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded mindustry-copper task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "integration_state": "validated",
      "preview_url": "https://store.steampowered.com/app/1127400/",
      "preview_label": "Watch official trailer",
      "task_ids": [
        "mindustry-copper"
      ],
      "video": "media/mindustry.mp4",
      "animation": "media/mindustry.gif",
      "preview_task": "mindustry-copper"
    },
    {
      "name": "OpenTTD",
      "category": "OPEN SOURCE \u00b7 2D SIMULATION",
      "note": "Build the first owned road on a seeded 64\u00d764 map using the normal toolbar and mouse. Native road ownership, construction and exact delayed replay are verified. This bounded task is outside the frozen v0.2 model standings.",
      "url": "https://github.com/OpenTTD/OpenTTD",
      "status": "runnable",
      "id": "openttd",
      "description": "Build a transport company using trains, roads, ships and aircraft. Profitable routes depend on network design, capacity and changing demand.",
      "image": "media/openttd.webp",
      "image_alt": "OpenTTD: recorded openttd-first-road gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.3.0",
      "media_credit": "GeneralGameBench capture; OpenTTD",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded openttd-first-road task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "integration_state": "validated",
      "preview_url": "https://store.steampowered.com/app/1536610/",
      "preview_label": "Watch official trailer",
      "task_ids": [
        "openttd-first-road"
      ],
      "video": "media/openttd.mp4",
      "animation": "media/openttd.gif",
      "preview_task": "openttd-first-road"
    },
    {
      "name": "NetHack",
      "category": "ROGUELIKE \u00b7 2D",
      "note": "Native replay passed on macOS and Linux x86_64: nethack-score. Explore NetHack as the player character, survive and gain game score. Compass actions move or attack. The image is the game's native tile renderer. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/facebookresearch/nle",
      "status": "runnable",
      "id": "nethack",
      "description": "Explore a dangerous dungeon filled with monsters, objects and uncertain consequences. Turn-based decisions reward careful resource use and planning.",
      "image": "media/nethack.webp",
      "image_alt": "NetHack: recorded nethack-score gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "NetHack and NLE contributors",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded nethack-score task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "video": "media/nethack.mp4",
      "animation": "media/nethack.gif",
      "task_ids": [
        "nethack-score"
      ],
      "integration_state": "validated",
      "preview_task": "nethack-score"
    },
    {
      "name": "MiniHack",
      "category": "RESEARCH \u00b7 2D",
      "note": "Native replay passed on macOS and Linux x86_64: minihack-room. Navigate the small room to the staircase. Use compass directions; the goal gives reward one. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/facebookresearch/minihack",
      "status": "runnable",
      "id": "minihack",
      "description": "Small NetHack-based worlds isolate skills such as navigation, tool use and combat. The tiles show compact problems with different goals.",
      "image": "media/minihack.webp",
      "image_alt": "MiniHack: recorded minihack-room gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "MiniHack / Meta research contributors",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded minihack-room task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "video": "media/minihack.mp4",
      "animation": "media/minihack.gif",
      "task_ids": [
        "minihack-room"
      ],
      "integration_state": "validated",
      "preview_task": "minihack-room"
    },
    {
      "name": "Dungeon Crawl Stone Soup",
      "category": "ROGUELIKE \u00b7 2D",
      "note": "Native tiles, Minotaur Fighter, Dungeon:1. Earn the first experience point through ordinary movement/combat keys; native reward, death and exact replay are verified. This bounded task is outside the frozen v0.2 model standings.",
      "url": "https://github.com/crawl/crawl",
      "status": "runnable",
      "id": "dungeon-crawl-stone-soup",
      "description": "Descend through a turn-based dungeon and search for the Orb of Zot. Character abilities, positioning and scarce resources shape each encounter.",
      "image": "media/dungeon-crawl-stone-soup.webp",
      "image_alt": "Dungeon Crawl Stone Soup: recorded dcss-first-experience gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.3.0",
      "media_credit": "Dungeon Crawl Stone Soup contributors",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded dcss-first-experience task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "integration_state": "validated",
      "preview_url": "https://crawl.develz.org/wordpress/screenshots",
      "preview_label": "View game gallery",
      "task_ids": [
        "dcss-first-experience"
      ],
      "video": "media/dungeon-crawl-stone-soup.mp4",
      "animation": "media/dungeon-crawl-stone-soup.gif",
      "preview_task": "dcss-first-experience"
    },
    {
      "name": "Cataclysm: Dark Days Ahead",
      "category": "SURVIVAL \u00b7 2D",
      "note": "Native tutorial with UltimateCataclysm tiles: find and wield the baseball bat using ordinary keys. Native equipment scoring and exact delayed replay are verified. This bounded task is outside the frozen v0.2 model standings.",
      "url": "https://github.com/CleverRaven/Cataclysm-DDA",
      "status": "runnable",
      "id": "cataclysm-dark-days-ahead",
      "description": "Survive in a ruined world by scavenging, crafting and avoiding threats. Detailed inventories and changing needs make long-term decisions matter.",
      "image": "media/cataclysm-dark-days-ahead.webp",
      "image_alt": "Cataclysm: Dark Days Ahead: recorded cdda-first-weapon gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.3.0",
      "media_credit": "Cataclysm: DDA contributors / Ultica artists",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded cdda-first-weapon task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "integration_state": "validated",
      "preview_url": "https://github.com/CleverRaven/Cataclysm-DDA",
      "preview_label": "View project media",
      "task_ids": [
        "cdda-first-weapon"
      ],
      "video": "media/cataclysm-dark-days-ahead.mp4",
      "animation": "media/cataclysm-dark-days-ahead.gif",
      "preview_task": "cdda-first-weapon"
    },
    {
      "name": "Crafter",
      "category": "RESEARCH \u00b7 2D SURVIVAL",
      "note": "Native replay passed on macOS and Linux x86_64: crafter. Survive, collect resources, craft tools and unlock achievements. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/danijar/crafter",
      "status": "runnable",
      "id": "crafter",
      "description": "Gather resources, craft tools and survive in a small procedural world. Progress is measured through achievements rather than one final objective.",
      "image": "media/crafter.webp",
      "image_alt": "Crafter: recorded crafter gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "Danijar Hafner / Crafter contributors",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded crafter task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "video": "media/crafter.mp4",
      "animation": "media/crafter.gif",
      "task_ids": [
        "crafter"
      ],
      "integration_state": "validated",
      "preview_task": "crafter"
    },
    {
      "name": "MiniWorld",
      "category": "RESEARCH \u00b7 3D",
      "note": "Native replay passed on macOS and Linux x86_64: miniworld-oneroom. Navigate the first-person room and touch the red box. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/Farama-Foundation/Miniworld",
      "status": "runnable",
      "id": "miniworld",
      "description": "Navigate simple three-dimensional rooms, paths and mazes from a first-person view. Compact tasks make spatial perception and movement easy to study.",
      "image": "media/miniworld.webp",
      "image_alt": "MiniWorld: recorded miniworld-oneroom gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "Farama Foundation / MiniWorld contributors",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded miniworld-oneroom task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "task_ids": [
        "miniworld-oneroom"
      ],
      "integration_state": "validated",
      "video": "media/miniworld.mp4",
      "animation": "media/miniworld.gif",
      "preview_task": "miniworld-oneroom"
    },
    {
      "name": "Unity ML-Agents environments",
      "category": "MODERN ENGINE \u00b7 3D / 2D",
      "note": "Validated VisualFoodCollector scene with the official pinned executables for macOS and Linux x86_64. Native food reward, goal collision and exact image/score replay passed; private vectors stay referee-only. This one scene is not in the frozen v0.2 model standings.",
      "url": "https://github.com/Unity-Technologies/ml-agents",
      "status": "runnable",
      "id": "unity-ml-agents-environments",
      "description": "A collection of Unity scenes explores balance, locomotion, navigation and team play. Each scene defines its own observations, actions and goals.",
      "image": "media/unity-ml-agents-environments.webp",
      "image_alt": "Unity ML-Agents environments: recorded unity-food-collector gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.3.0",
      "media_credit": "Unity Technologies / ML-Agents contributors",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded unity-food-collector task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "integration_state": "validated",
      "task_ids": [
        "unity-food-collector"
      ],
      "video": "media/unity-ml-agents-environments.mp4",
      "animation": "media/unity-ml-agents-environments.gif",
      "preview_task": "unity-food-collector"
    },
    {
      "name": "Google Research Football",
      "category": "RESEARCH \u00b7 3D SPORT",
      "note": "Validated Football Academy close-range goal task in Linux ARM64 and x86_64. Native goal scoring, termination and exact image/score replay passed. This Academy scenario is not a full football match or part of the frozen v0.2 model standings.",
      "url": "https://github.com/google-research/football",
      "status": "runnable",
      "id": "google-research-football",
      "description": "Control football players in a three-dimensional match simulation. Passing, positioning and shots combine quick control with team tactics.",
      "image": "media/google-research-football.webp",
      "image_alt": "Google Research Football: recorded football-empty-goal gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.3.0",
      "media_credit": "Google Research / Football contributors",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded football-empty-goal task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "video": "media/google-research-football.mp4",
      "animation": "media/google-research-football.gif",
      "integration_state": "validated",
      "task_ids": [
        "football-empty-goal"
      ],
      "preview_task": "football-empty-goal"
    },
    {
      "name": "Minecraft",
      "category": "COMMERCIAL \u00b7 3D SANDBOX",
      "note": "MineRL/Malmo are established research routes, with version and installation constraints. Requires a lawful game setup and adapter. No usable local installation was available for this integration pass.",
      "url": "https://github.com/minerllabs/minerl",
      "status": "candidate",
      "id": "minecraft",
      "description": "Explore a block-built world, gather materials and craft increasingly complex tools. Open-ended survival creates long sequences of dependent goals.",
      "image": "media/minecraft.webp",
      "image_alt": "Minecraft gameplay preview",
      "media_source": "https://minerl.readthedocs.io/en/latest/",
      "media_credit": "MineRL demonstration; Minecraft by Mojang Studios",
      "media_kind": "Project preview",
      "media_caption": "MineRL gameplay example; no Minecraft benchmark has been run here.",
      "video": "media/minecraft.mp4",
      "animation": "media/minecraft.gif",
      "integration_state": "assets-required"
    },
    {
      "name": "StarCraft II",
      "category": "AAA \u00b7 3D STRATEGY",
      "note": "Official Linux research client runs on this Mac with native RGB and ordinary selection/minimap controls. MoveToBeacon scores and replays game progress, but animated pixels differ. The diagnostic is not ranked.",
      "url": "https://github.com/Blizzard/s2client-api",
      "status": "candidate",
      "id": "starcraft-ii",
      "description": "Build a base, gather resources and command armies in real-time battles. Strategy, scouting and precise unit control all affect the outcome.",
      "image": "media/starcraft-ii.webp",
      "image_alt": "StarCraft II: native diagnostic capture, unranked",
      "media_source": "https://github.com/DanielTea/generalgamebench/tree/v0.3.0/environments/experimental-sc2",
      "media_credit": "GeneralGameBench capture; Blizzard Entertainment / DeepMind mini-game maps",
      "media_kind": "Native diagnostic \u00b7 unranked",
      "media_caption": "Native prototype capture. Exact visual replay failed; this candidate has no admitted task or ranking. Consecutive frames at 6 fps do not measure reaction time.",
      "integration_state": "experimental",
      "preview_url": "https://starcraft2.blizzard.com/en-us/",
      "preview_label": "View project media",
      "video": "media/starcraft-ii.mp4",
      "animation": "media/starcraft-ii.gif"
    },
    {
      "name": "Rocket League / RocketSim",
      "category": "COMMERCIAL \u00b7 3D SPORT",
      "note": "RLGym is a research route. Simulator and retail-game results must be distinct; visual capture and access constraints need validation. No usable local installation was available for this integration pass.",
      "url": "https://github.com/RLGym/rlgym",
      "status": "candidate",
      "id": "rocket-league-rocketsim",
      "description": "Drive rocket-powered cars to knock a giant ball into the opposing goal. Steering, aerial movement and team positioning meet in fast matches.",
      "image": "media/rocket-league-rocketsim.webp",
      "image_alt": "Rocket League / RocketSim gameplay preview",
      "media_source": "https://store.steampowered.com/app/252950/",
      "media_credit": "Psyonix LLC / publisher-provided Steam screenshot",
      "media_kind": "Project preview",
      "media_caption": "Representative game preview; integration remains a research candidate.",
      "integration_state": "assets-required",
      "preview_url": "https://store.steampowered.com/app/252950/",
      "preview_label": "Watch official trailer"
    },
    {
      "name": "Factorio",
      "category": "INDIE \u00b7 2D AUTOMATION",
      "note": "Official Lua API can implement a trusted referee. Agents would receive screenshots only. Game license and deterministic scenario needed. No usable local installation was available for this integration pass.",
      "url": "https://lua-api.factorio.com/latest/",
      "status": "candidate",
      "id": "factorio",
      "description": "Design a factory that turns raw resources into increasingly complex products. Transport networks, power and expanding production form a large planning puzzle.",
      "image": "media/factorio.webp",
      "image_alt": "Factorio gameplay preview",
      "media_source": "https://store.steampowered.com/app/427520/",
      "media_credit": "Wube Software LTD. / publisher-provided Steam screenshot",
      "media_kind": "Project preview",
      "media_caption": "Representative game preview; integration remains a research candidate.",
      "integration_state": "assets-required",
      "preview_url": "https://store.steampowered.com/app/427520/",
      "preview_label": "Watch official trailer"
    },
    {
      "name": "Cyberpunk 2077",
      "category": "AAA \u00b7 MODERN 3D",
      "note": "Official REDmod exists. A repeatable offline benchmark is exploratory, requires licensed installs, hardware and task/referee work. No usable local installation was available for this integration pass.",
      "url": "https://www.cyberpunk.net/en/modding-support",
      "status": "candidate",
      "id": "cyberpunk-2077",
      "description": "Explore Night City in a first-person action role-playing game. Missions combine navigation, dialogue, combat and different approaches to a goal.",
      "image": "media/cyberpunk-2077.webp",
      "image_alt": "Cyberpunk 2077 gameplay preview",
      "media_source": "https://store.steampowered.com/app/1091500/",
      "media_credit": "CD PROJEKT RED / publisher-provided Steam screenshot",
      "media_kind": "Project preview",
      "media_caption": "Representative game preview; integration remains a research candidate.",
      "integration_state": "assets-required",
      "preview_url": "https://store.steampowered.com/app/1091500/",
      "preview_label": "Watch official trailer"
    },
    {
      "name": "Morrowind / OpenMW",
      "category": "COMMERCIAL ASSETS \u00b7 3D RPG",
      "note": "Open-source engine is a viable integration route; original game assets require separate rights. Scripted tasks and capture not implemented. No usable local installation was available for this integration pass.",
      "url": "https://github.com/OpenMW/openmw",
      "status": "candidate",
      "id": "morrowind-openmw",
      "description": "Explore Morrowind through the open-source OpenMW engine. Quests, dialogue, navigation and combat offer different kinds of decision making.",
      "image": "media/morrowind-openmw.webp",
      "image_alt": "Morrowind / OpenMW gameplay preview",
      "media_source": "https://openmw.org/media/",
      "media_credit": "OpenMW community; Morrowind assets by Bethesda Softworks",
      "media_kind": "Project preview",
      "media_caption": "Representative game preview; integration remains a research candidate.",
      "integration_state": "assets-required",
      "preview_url": "https://openmw.org/media/",
      "preview_label": "View game gallery"
    },
    {
      "name": "Half-Life 2 / Source SDK",
      "category": "AAA \u00b7 3D FPS",
      "note": "Source SDK offers an offline research path. Game assets and SDK terms remain separate; full adapter and validation needed. No usable local installation was available for this integration pass.",
      "url": "https://github.com/ValveSoftware/source-sdk-2013",
      "status": "candidate",
      "id": "half-life-2-source-sdk",
      "description": "Move through a first-person adventure with combat and physics puzzles. A gravity gun turns objects in the environment into tools and weapons.",
      "image": "media/half-life-2-source-sdk.webp",
      "image_alt": "Half-Life 2 / Source SDK gameplay preview",
      "media_source": "https://store.steampowered.com/app/220/",
      "media_credit": "Valve / publisher-provided Steam screenshot",
      "media_kind": "Project preview",
      "media_caption": "Representative game preview; integration remains a research candidate.",
      "integration_state": "assets-required",
      "preview_url": "https://store.steampowered.com/app/220/",
      "preview_label": "Watch official trailer"
    },
    {
      "name": "Atari / Arcade Learning Environment",
      "category": "RETRO \u00b7 2D FAMILY",
      "note": "Native replay passed on macOS and Linux x86_64: atari-breakout. Move the paddle and break bricks. Fire starts the ball. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/Farama-Foundation/Arcade-Learning-Environment",
      "status": "runnable",
      "id": "atari-arcade-learning-environment",
      "description": "A collection of classic Atari games spans ball control, shooting and navigation. Simple-looking screens can demand precise timing and sustained control.",
      "image": "media/atari-arcade-learning-environment.webp",
      "image_alt": "Atari / Arcade Learning Environment: recorded atari-breakout gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "Farama Foundation; Breakout by Atari",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded atari-breakout task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "video": "media/atari-arcade-learning-environment.mp4",
      "animation": "media/atari-arcade-learning-environment.gif",
      "task_ids": [
        "atari-breakout"
      ],
      "integration_state": "validated",
      "preview_task": "atari-breakout"
    },
    {
      "name": "Stable-Retro library",
      "category": "RETRO \u00b7 2D FAMILY",
      "note": "Native replay passed on macOS and Linux x86_64: retro-airstriker. Pilot the ship, dodge enemies and shoot. This is the freely distributed Airstriker Genesis demo. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/Farama-Foundation/stable-retro",
      "status": "runnable",
      "id": "stable-retro-library",
      "description": "An emulator-based collection connects classic console games to agent interfaces. Different titles bring platforming, racing, fighting and other control tasks.",
      "image": "media/stable-retro-library.webp",
      "image_alt": "Stable-Retro library: recorded retro-airstriker gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "Airstriker demo authors / Stable-Retro contributors",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded retro-airstriker task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "task_ids": [
        "retro-airstriker"
      ],
      "integration_state": "validated",
      "video": "media/stable-retro-library.mp4",
      "animation": "media/stable-retro-library.gif",
      "preview_task": "retro-airstriker"
    },
    {
      "name": "PettingZoo environments",
      "category": "MULTI-AGENT \u00b7 2D FAMILY",
      "note": "Native replay passed on macOS and Linux x86_64: pettingzoo-pistonball. Move the ball left. Control the middle piston; the other pistons follow a fixed periodic policy. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/Farama-Foundation/PettingZoo",
      "status": "runnable",
      "id": "pettingzoo-environments",
      "description": "A collection of cooperative and competitive multi-agent games. Agents may need to coordinate, anticipate opponents or share control of a task.",
      "image": "media/pettingzoo-environments.webp",
      "image_alt": "PettingZoo environments: recorded pettingzoo-pistonball gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "Farama Foundation / PettingZoo contributors",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded pettingzoo-pistonball task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "video": "media/pettingzoo-environments.mp4",
      "animation": "media/pettingzoo-environments.gif",
      "task_ids": [
        "pettingzoo-pistonball"
      ],
      "integration_state": "validated",
      "preview_task": "pettingzoo-pistonball"
    },
    {
      "name": "Procgen / BigFish",
      "category": "PROCEDURAL \u00b7 2D",
      "note": "Native replay passed on macOS and Linux x86_64: procgen-bigfish. Eat smaller fish and avoid larger fish. Move in four directions. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/openai/procgen",
      "status": "runnable",
      "id": "procgen-bigfish",
      "description": "Eat smaller fish, grow larger and keep clear of bigger predators.",
      "image": "media/procgen-bigfish.webp",
      "image_alt": "Procgen / BigFish: recorded procgen-bigfish gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "OpenAI / Procgen contributors; assets retain upstream licenses",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded procgen-bigfish task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "task_ids": [
        "procgen-bigfish"
      ],
      "integration_state": "validated",
      "video": "media/procgen-bigfish.mp4",
      "animation": "media/procgen-bigfish.gif",
      "preview_task": "procgen-bigfish"
    },
    {
      "name": "Procgen / BossFight",
      "category": "PROCEDURAL \u00b7 2D",
      "note": "Native replay passed on macOS and Linux x86_64: procgen-bossfight. Dodge enemy projectiles and shoot the boss. D fires. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/openai/procgen",
      "status": "runnable",
      "id": "procgen-bossfight",
      "description": "Dodge a boss ship\u2019s projectiles and attack when its shield drops.",
      "image": "media/procgen-bossfight.webp",
      "image_alt": "Procgen / BossFight: recorded procgen-bossfight gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "OpenAI / Procgen contributors; assets retain upstream licenses",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded procgen-bossfight task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "task_ids": [
        "procgen-bossfight"
      ],
      "integration_state": "validated",
      "video": "media/procgen-bossfight.mp4",
      "animation": "media/procgen-bossfight.gif",
      "preview_task": "procgen-bossfight"
    },
    {
      "name": "Procgen / CaveFlyer",
      "category": "PROCEDURAL \u00b7 2D",
      "note": "Native replay passed on macOS and Linux x86_64: procgen-caveflyer. Fly through the cave to the exit without hitting walls. D fires. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/openai/procgen",
      "status": "runnable",
      "id": "procgen-caveflyer",
      "description": "Pilot a rotating ship through dangerous caves to reach the exit.",
      "image": "media/procgen-caveflyer.webp",
      "image_alt": "Procgen / CaveFlyer: recorded procgen-caveflyer gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "OpenAI / Procgen contributors; assets retain upstream licenses",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded procgen-caveflyer task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "task_ids": [
        "procgen-caveflyer"
      ],
      "integration_state": "validated",
      "video": "media/procgen-caveflyer.mp4",
      "animation": "media/procgen-caveflyer.gif",
      "preview_task": "procgen-caveflyer"
    },
    {
      "name": "Procgen / Chaser",
      "category": "PROCEDURAL \u00b7 2D",
      "note": "Native replay passed on macOS and Linux x86_64: procgen-chaser. Collect green orbs and avoid chasing enemies. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/openai/procgen",
      "status": "runnable",
      "id": "procgen-chaser",
      "description": "Collect maze orbs while avoiding enemies and using temporary power-ups.",
      "image": "media/procgen-chaser.webp",
      "image_alt": "Procgen / Chaser: recorded procgen-chaser gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "OpenAI / Procgen contributors; assets retain upstream licenses",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded procgen-chaser task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "task_ids": [
        "procgen-chaser"
      ],
      "integration_state": "validated",
      "video": "media/procgen-chaser.mp4",
      "animation": "media/procgen-chaser.gif",
      "preview_task": "procgen-chaser"
    },
    {
      "name": "Procgen / Climber",
      "category": "PROCEDURAL \u00b7 2D",
      "note": "Native replay passed on macOS and Linux x86_64: procgen-climber. Climb platforms and collect stars; up jumps. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/openai/procgen",
      "status": "runnable",
      "id": "procgen-climber",
      "description": "Jump between platforms, gather stars and stay clear of flying enemies.",
      "image": "media/procgen-climber.webp",
      "image_alt": "Procgen / Climber: recorded procgen-climber gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "OpenAI / Procgen contributors; assets retain upstream licenses",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded procgen-climber task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "task_ids": [
        "procgen-climber"
      ],
      "integration_state": "validated",
      "video": "media/procgen-climber.mp4",
      "animation": "media/procgen-climber.gif",
      "preview_task": "procgen-climber"
    },
    {
      "name": "Procgen / CoinRun",
      "category": "PROCEDURAL \u00b7 2D",
      "note": "Native replay passed on macOS and Linux x86_64: procgen-coinrun. Reach the coin at the end of the platform level; up jumps. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/openai/procgen",
      "status": "runnable",
      "id": "procgen-coinrun",
      "description": "Cross a platforming level, avoiding hazards to collect the final coin.",
      "image": "media/procgen-coinrun.webp",
      "image_alt": "Procgen / CoinRun: recorded procgen-coinrun gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "OpenAI / Procgen contributors; assets retain upstream licenses",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded procgen-coinrun task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "task_ids": [
        "procgen-coinrun"
      ],
      "integration_state": "validated",
      "video": "media/procgen-coinrun.mp4",
      "animation": "media/procgen-coinrun.gif",
      "preview_task": "procgen-coinrun"
    },
    {
      "name": "Procgen / Dodgeball",
      "category": "PROCEDURAL \u00b7 2D",
      "note": "Native replay passed on macOS and Linux x86_64: procgen-dodgeball. Avoid enemy balls and defeat enemies; D throws a ball. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/openai/procgen",
      "status": "runnable",
      "id": "procgen-dodgeball",
      "description": "Dodge enemy balls, return fire and reach the exit after clearing opponents.",
      "image": "media/procgen-dodgeball.webp",
      "image_alt": "Procgen / Dodgeball: recorded procgen-dodgeball gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "OpenAI / Procgen contributors; assets retain upstream licenses",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded procgen-dodgeball task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "task_ids": [
        "procgen-dodgeball"
      ],
      "integration_state": "validated",
      "video": "media/procgen-dodgeball.mp4",
      "animation": "media/procgen-dodgeball.gif",
      "preview_task": "procgen-dodgeball"
    },
    {
      "name": "Procgen / FruitBot",
      "category": "PROCEDURAL \u00b7 2D",
      "note": "Native replay passed on macOS and Linux x86_64: procgen-fruitbot. Collect fruit while avoiding obstacles and non-fruit objects; D fires. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/openai/procgen",
      "status": "runnable",
      "id": "procgen-fruitbot",
      "description": "Steer through gaps, collect fruit and avoid harmful objects while scrolling.",
      "image": "media/procgen-fruitbot.webp",
      "image_alt": "Procgen / FruitBot: recorded procgen-fruitbot gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "OpenAI / Procgen contributors; assets retain upstream licenses",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded procgen-fruitbot task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "task_ids": [
        "procgen-fruitbot"
      ],
      "integration_state": "validated",
      "video": "media/procgen-fruitbot.mp4",
      "animation": "media/procgen-fruitbot.gif",
      "preview_task": "procgen-fruitbot"
    },
    {
      "name": "Procgen / Heist",
      "category": "PROCEDURAL \u00b7 2D",
      "note": "Native replay passed on macOS and Linux x86_64: procgen-heist. Collect keys to open matching doors and reach the gem. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/openai/procgen",
      "status": "runnable",
      "id": "procgen-heist",
      "description": "Find colored keys and unlock a maze route to the hidden gem.",
      "image": "media/procgen-heist.webp",
      "image_alt": "Procgen / Heist: recorded procgen-heist gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "OpenAI / Procgen contributors; assets retain upstream licenses",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded procgen-heist task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "task_ids": [
        "procgen-heist"
      ],
      "integration_state": "validated",
      "video": "media/procgen-heist.mp4",
      "animation": "media/procgen-heist.gif",
      "preview_task": "procgen-heist"
    },
    {
      "name": "Procgen / Jumper",
      "category": "PROCEDURAL \u00b7 2D",
      "note": "Native replay passed on macOS and Linux x86_64: procgen-jumper. Jump across platforms to the carrot; up jumps. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/openai/procgen",
      "status": "runnable",
      "id": "procgen-jumper",
      "description": "Use double jumps to find a carrot in an open platforming level.",
      "image": "media/procgen-jumper.webp",
      "image_alt": "Procgen / Jumper: recorded procgen-jumper gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "OpenAI / Procgen contributors; assets retain upstream licenses",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded procgen-jumper task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "task_ids": [
        "procgen-jumper"
      ],
      "integration_state": "validated",
      "video": "media/procgen-jumper.mp4",
      "animation": "media/procgen-jumper.gif",
      "preview_task": "procgen-jumper"
    },
    {
      "name": "Procgen / Leaper",
      "category": "PROCEDURAL \u00b7 2D",
      "note": "Native replay passed on macOS and Linux x86_64: procgen-leaper. Cross roads and water to reach the finish without being hit. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/openai/procgen",
      "status": "runnable",
      "id": "procgen-leaper",
      "description": "Cross traffic and hop between floating logs to reach the far side.",
      "image": "media/procgen-leaper.webp",
      "image_alt": "Procgen / Leaper: recorded procgen-leaper gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "OpenAI / Procgen contributors; assets retain upstream licenses",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded procgen-leaper task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "task_ids": [
        "procgen-leaper"
      ],
      "integration_state": "validated",
      "video": "media/procgen-leaper.mp4",
      "animation": "media/procgen-leaper.gif",
      "preview_task": "procgen-leaper"
    },
    {
      "name": "Procgen / Maze",
      "category": "PROCEDURAL \u00b7 2D",
      "note": "Native replay passed on macOS and Linux x86_64: procgen-maze. Navigate the maze to the cheese using four-direction movement. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/openai/procgen",
      "status": "runnable",
      "id": "procgen-maze",
      "description": "Guide a mouse through a generated maze to find its cheese.",
      "image": "media/procgen-maze.webp",
      "image_alt": "Procgen / Maze: recorded procgen-maze gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "OpenAI / Procgen contributors; assets retain upstream licenses",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded procgen-maze task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "task_ids": [
        "procgen-maze"
      ],
      "integration_state": "validated",
      "video": "media/procgen-maze.mp4",
      "animation": "media/procgen-maze.gif",
      "preview_task": "procgen-maze"
    },
    {
      "name": "Procgen / Miner",
      "category": "PROCEDURAL \u00b7 2D",
      "note": "Native replay passed on macOS and Linux x86_64: procgen-miner. Collect diamonds, avoid falling rocks, and reach the exit. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/openai/procgen",
      "status": "runnable",
      "id": "procgen-miner",
      "description": "Dig for diamonds while avoiding falling rocks, then find the exit.",
      "image": "media/procgen-miner.webp",
      "image_alt": "Procgen / Miner: recorded procgen-miner gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "OpenAI / Procgen contributors; assets retain upstream licenses",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded procgen-miner task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "task_ids": [
        "procgen-miner"
      ],
      "integration_state": "validated",
      "video": "media/procgen-miner.mp4",
      "animation": "media/procgen-miner.gif",
      "preview_task": "procgen-miner"
    },
    {
      "name": "Procgen / Ninja",
      "category": "PROCEDURAL \u00b7 2D",
      "note": "Native replay passed on macOS and Linux x86_64: procgen-ninja. Reach the mushroom; hold up to charge a jump and release to jump. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/openai/procgen",
      "status": "runnable",
      "id": "procgen-ninja",
      "description": "Charge jumps across narrow ledges and clear bombs with throwing stars.",
      "image": "media/procgen-ninja.webp",
      "image_alt": "Procgen / Ninja: recorded procgen-ninja gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "OpenAI / Procgen contributors; assets retain upstream licenses",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded procgen-ninja task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "task_ids": [
        "procgen-ninja"
      ],
      "integration_state": "validated",
      "video": "media/procgen-ninja.mp4",
      "animation": "media/procgen-ninja.gif",
      "preview_task": "procgen-ninja"
    },
    {
      "name": "Procgen / Plunder",
      "category": "PROCEDURAL \u00b7 2D",
      "note": "Native replay passed on macOS and Linux x86_64: procgen-plunder. Shoot enemy pirate ships and avoid friendly ships; D fires. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/openai/procgen",
      "status": "runnable",
      "id": "procgen-plunder",
      "description": "Fire cannons at hostile pirate ships while sparing friendly vessels.",
      "image": "media/procgen-plunder.webp",
      "image_alt": "Procgen / Plunder: recorded procgen-plunder gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "OpenAI / Procgen contributors; assets retain upstream licenses",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded procgen-plunder task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "task_ids": [
        "procgen-plunder"
      ],
      "integration_state": "validated",
      "video": "media/procgen-plunder.mp4",
      "animation": "media/procgen-plunder.gif",
      "preview_task": "procgen-plunder"
    },
    {
      "name": "Procgen / StarPilot",
      "category": "PROCEDURAL \u00b7 2D",
      "note": "Native replay passed on macOS and Linux x86_64: procgen-starpilot. Fly, dodge hazards and shoot enemies; D fires. Exact frame and score replay passed. This validates these tasks only, not every game in a library.",
      "url": "https://github.com/openai/procgen",
      "status": "runnable",
      "id": "procgen-starpilot",
      "description": "Pilot a scrolling spaceship through enemies, aimed projectiles and obstacles.",
      "image": "media/procgen-starpilot.webp",
      "image_alt": "Procgen / StarPilot: recorded procgen-starpilot gameplay",
      "media_source": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.2.0",
      "media_credit": "OpenAI / Procgen contributors; assets retain upstream licenses",
      "media_kind": "Recorded benchmark",
      "media_caption": "Recorded procgen-starpilot task with the random reference policy. Consecutive evaluation frames at 6 fps; playback speed does not measure reaction time.",
      "task_ids": [
        "procgen-starpilot"
      ],
      "integration_state": "validated",
      "video": "media/procgen-starpilot.mp4",
      "animation": "media/procgen-starpilot.gif",
      "preview_task": "procgen-starpilot"
    }
  ],
  "coverage": {
    "task_count": 43,
    "validated_cards": 26,
    "catalog_cards": 45
  },
  "model_status": [
    {
      "agent": "mlx-mlx-community-SmolVLM2-256M-Video-Instruct-mlx",
      "provider": "mlx",
      "model": "mlx-community/SmolVLM2-256M-Video-Instruct-mlx",
      "revision": "79901655c1a3d7ed6646c91325e3fde47e40e611",
      "status": "complete",
      "completed": 43
    },
    {
      "agent": "mlx-mlx-community-SmolVLM2-500M-Video-Instruct-mlx",
      "provider": "mlx",
      "model": "mlx-community/SmolVLM2-500M-Video-Instruct-mlx",
      "revision": "fa57db46815177fbdfd65cc85a2b3416a8332268",
      "status": "complete",
      "completed": 43
    },
    {
      "agent": "mlx-LiquidAI-LFM2-5-VL-450M-MLX-4bit",
      "provider": "mlx",
      "model": "LiquidAI/LFM2.5-VL-450M-MLX-4bit",
      "revision": "f19926f17a25164d4cbcdc16d9eaf4714b807cfb",
      "status": "complete",
      "completed": 43
    },
    {
      "agent": "mlx-mlx-community-FastVLM-0-5B-bf16",
      "provider": "mlx",
      "model": "mlx-community/FastVLM-0.5B-bf16",
      "revision": "81ffe929046666c43de53691147b1669ba0f3a4c",
      "status": "complete",
      "completed": 43
    },
    {
      "agent": "claude-claude-sonnet-5-5",
      "provider": "claude",
      "model": "claude-sonnet-5-5",
      "revision": null,
      "status": "complete",
      "completed": 43
    },
    {
      "agent": "openai-gpt-6-astra",
      "provider": "openai",
      "model": "gpt-6-astra",
      "revision": null,
      "status": "complete",
      "completed": 43
    },
    {
      "agent": "openai-gpt-6-sol",
      "provider": "openai",
      "model": "gpt-6-sol",
      "revision": null,
      "status": "complete",
      "completed": 43
    },
    {
      "agent": "openai-gpt-6-luna",
      "provider": "openai",
      "model": "gpt-6-luna",
      "revision": null,
      "status": "complete",
      "completed": 43
    },
    {
      "agent": "openai-gpt-5-6-sol",
      "provider": "openai",
      "model": "gpt-5.6-sol",
      "revision": null,
      "status": "complete",
      "completed": 43
    },
    {
      "agent": "openai-gpt-5-6-terra",
      "provider": "openai",
      "model": "gpt-5.6-terra",
      "revision": null,
      "status": "complete",
      "completed": 43
    },
    {
      "agent": "openai-gpt-5-6-luna",
      "provider": "openai",
      "model": "gpt-5.6-luna",
      "revision": null,
      "status": "complete",
      "completed": 43
    },
    {
      "agent": "openai-gpt-5-5",
      "provider": "openai",
      "model": "gpt-5.5",
      "revision": null,
      "status": "complete",
      "completed": 43
    },
    {
      "agent": "claude-claude-opus-5-1m-",
      "provider": "claude",
      "model": "claude-opus-5[1m]",
      "revision": null,
      "status": "complete",
      "completed": 43
    },
    {
      "agent": "claude-claude-sonnet-5",
      "provider": "claude",
      "model": "claude-sonnet-5",
      "revision": null,
      "status": "complete",
      "completed": 43
    },
    {
      "agent": "claude-claude-haiku-4-5-20251001",
      "provider": "claude",
      "model": "claude-haiku-4-5-20251001",
      "revision": null,
      "status": "complete",
      "completed": 43
    }
  ],
  "hardware_details": "Apple M3 Max; 48 GiB RAM; 16 CPU cores; macOS arm64. One local MLX model and one game run at a time. Weights remain loaded between games. Each model receives three gray startup images before scored actions. Startup image sizes: 160x120, 320x240, 960x640. Startup has no game data. Every game action remains in the measured result. Normal desktop tasks can use the same Mac.",
  "suite_files": [
    "benchmarks/mac-hosted-rerun-2026-10-05.json",
    "benchmarks/mac-hf-vlm-2026-10-05.json",
    "benchmarks/mac-baselines-2026-10-05.json",
    "benchmarks/mac-hosted-exhibition-2026-10-05.json",
    "benchmarks/mac-exhibition-v1.json",
    "benchmarks/mac-baselines-v1.json"
  ],
  "integration_coverage": {
    "version": "0.3.0",
    "task_count": 43,
    "validated_card_count": 36,
    "catalog_card_count": 45,
    "unadmitted_card_count": 9,
    "standings_task_count": 33,
    "standings_version": "0.2.0",
    "new_tasks": [
      "unity-food-collector",
      "football-empty-goal",
      "supertuxkart-lighthouse",
      "luanti-chop-tree",
      "supertux-first-coin",
      "dcss-first-experience",
      "openttd-first-road",
      "mindustry-copper",
      "cdda-first-weapon",
      "warzone-first-derrick"
    ],
    "evidence": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.3.0",
    "purpose": "New task admission; model standings remain the original v0.2 cohort.",
    "validation": {
      "episodes": 21,
      "decisions": 4303,
      "replayed": 21,
      "errors": 0,
      "aborted": 0,
      "model_calls": 0,
      "purpose": "integration-validation-only",
      "trust": "local-unattested",
      "release": "https://github.com/DanielTea/generalgamebench/releases/tag/v0.3.0"
    }
  },
  "track_metadata": {
    "exhibition": {
      "season": "hosted-rerun-2026-10-05",
      "hardware_details": "Game scores use the same fixed task definitions and seeds. Timing includes the full response path. Hosted CLI calls and persistent local MLX calls have different inference paths. Do not interpret this table as a model-only speed comparison."
    }
  },
  "current_campaign": "results/hosted-rerun-2026-10-05/campaign.json",
  "baseline_campaign": "results/baselines-2026-10-05/campaign.json",
  "latency_policy": {
    "id": "suite-p95-200-v1",
    "metric": "suite_p95_ms",
    "limit_ms": 200.0,
    "comparison": "strictly_less_than",
    "quantile_method": "linear",
    "sample_scope": "all_recorded_decisions_in_complete_suite"
  },
  "archives": [
    {
      "snapshot": "results/hf-vlm-2026-10-05/previous-snapshot.json",
      "sha256": "c029b38dc95d3a10d4390689610c589f15162048843ba2cd152535adb7ae116d",
      "season": "baselines-2026-10-05",
      "current_campaign": "results/refresh-2026-10-05/campaign.json",
      "note": "Keep the previous results and recording rules. The model table shows the new verified runs."
    },
    {
      "snapshot": "results/hosted-rerun-2026-10-05/previous-snapshot.json",
      "sha256": "83b8ac55725e66ed2db376b4ea651262d970b079fdd3b755dd32374b47b7775a",
      "season": "hf-vlm-2026-10-05",
      "note": "The previous four-model snapshot remains unchanged. Its verified model rows remain in the current table."
    }
  ],
  "campaigns": [
    "results/refresh-2026-10-05/campaign.json",
    "results/hf-vlm-2026-10-05/campaign.json",
    "results/hosted-rerun-2026-10-05/campaign.json"
  ],
  "comparison_note": "Game scores use the same fixed task definitions and seeds. Timing includes the full response path. Hosted CLI calls and persistent local MLX calls have different inference paths. Do not interpret this table as a model-only speed comparison."
}
