{
  "schemaVersion": 1,
  "title": "Game-agent evaluation audit",
  "createdAt": "2026-10-02T08:13:03.934Z",
  "method": "Documentation inspection and arithmetic only; no model or game execution.",
  "sources": [
    {
      "id": "bfl-announcement",
      "url": "https://huggingface.co/blog/black-forest-labs/flux-3-action",
      "publisher": "Black Forest Labs",
      "accessedAt": "2026-10-02T08:13:03.934Z"
    },
    {
      "id": "bfl-games",
      "url": "https://docs.bfl.ai/flux_3/flux3_action_games",
      "publisher": "Black Forest Labs",
      "accessedAt": "2026-10-02T08:13:03.934Z"
    },
    {
      "id": "bfl-inference",
      "url": "https://docs.bfl.ai/flux_3/flux3_action_inference",
      "publisher": "Black Forest Labs",
      "accessedAt": "2026-10-02T08:13:03.934Z"
    },
    {
      "id": "bfl-base",
      "url": "https://huggingface.co/black-forest-labs/flux-3-action-base",
      "publisher": "Black Forest Labs",
      "accessedAt": "2026-10-02T08:13:03.934Z"
    },
    {
      "id": "bfl-code",
      "url": "https://github.com/black-forest-labs/flux-action",
      "publisher": "Black Forest Labs",
      "accessedAt": "2026-10-02T08:13:03.934Z"
    },
    {
      "id": "bfl-license",
      "url": "https://huggingface.co/black-forest-labs/flux-3-action-base/blob/main/LICENSE.md",
      "publisher": "Black Forest Labs",
      "accessedAt": "2026-10-02T08:13:03.934Z"
    },
    {
      "id": "farama-env",
      "url": "https://gymnasium.farama.org/api/env/",
      "publisher": "Farama Foundation",
      "accessedAt": "2026-10-02T08:13:03.934Z"
    },
    {
      "id": "farama-checker",
      "url": "https://gymnasium.farama.org/api/utils/#environment-checking",
      "publisher": "Farama Foundation",
      "accessedAt": "2026-10-02T08:13:03.934Z"
    }
  ],
  "audit": [
    {
      "id": "release-completeness",
      "documented": "Game runtime and evaluation components are not all linked or supplied.",
      "sourceIds": [
        "bfl-games"
      ],
      "requiredForOwnEvaluation": "Identify simulator, bot, recorder, player, evaluator, versions, and run manifests."
    },
    {
      "id": "demo-scope",
      "documented": "Reported game comparisons use one held-out seed per game for 60 seconds.",
      "sourceIds": [
        "bfl-announcement"
      ],
      "requiredForOwnEvaluation": "Predefine a held-out scenario suite and preserve every run, including failures."
    },
    {
      "id": "control-timing",
      "documented": "Game plan length is 32; documented execution horizons include 2 and 8 actions at 15 Hz.",
      "sourceIds": [
        "bfl-inference"
      ],
      "requiredForOwnEvaluation": "Record plan length, execution horizon, pause mode, observation age, and end-to-end latency."
    },
    {
      "id": "api-output",
      "documented": "Current inference API returns actions, without decoded video.",
      "sourceIds": [
        "bfl-inference"
      ],
      "requiredForOwnEvaluation": "Treat predicted-video display as separate implementation work."
    },
    {
      "id": "outcome-contract",
      "documented": "Gymnasium separates task termination from external truncation and documents separate action-space seeding.",
      "sourceIds": [
        "farama-env"
      ],
      "requiredForOwnEvaluation": "Store outcome reason and distinct environment/controller seeds."
    },
    {
      "id": "interface-check",
      "documented": "Farama recommends check_env after environment creation and in CI.",
      "sourceIds": [
        "farama-checker"
      ],
      "requiredForOwnEvaluation": "Check API compliance separately from game correctness and useful defect coverage."
    },
    {
      "id": "license-boundary",
      "documented": "Code and weights carry different licenses.",
      "sourceIds": [
        "bfl-code",
        "bfl-license"
      ],
      "requiredForOwnEvaluation": "Review the actual intended use under each applicable license."
    }
  ],
  "arithmetic": {
    "formula": "control_time_ms = executed_actions / control_hz * 1000",
    "control_hz": 15,
    "sourceIds": [
      "bfl-inference"
    ],
    "excludes": [
      "inference",
      "communication",
      "scheduling delays"
    ],
    "rows": [
      {
        "executed_actions": 2,
        "control_time_ms": 133.33333333333334,
        "display_ms": "133.33"
      },
      {
        "executed_actions": 8,
        "control_time_ms": 533.3333333333334,
        "display_ms": "533.33"
      },
      {
        "executed_actions": 32,
        "control_time_ms": 2133.3333333333335,
        "display_ms": "2,133.33"
      }
    ],
    "reproduce_python": "[(n, n / 15 * 1000) for n in (2, 8, 32)]"
  },
  "blankRunRecord": {
    "status": "not-run",
    "gameBuild": null,
    "checkpoint": null,
    "scenarioId": null,
    "environmentSeed": null,
    "controllerSeed": null,
    "taskInstruction": null,
    "adapterVersion": null,
    "actionDecoderVersion": null,
    "hardware": null,
    "controlHz": null,
    "planLength": null,
    "executionHorizon": null,
    "pausesForInference": null,
    "warmupProtocol": null,
    "frameToCommandLatencyMs": null,
    "observationAgeMs": null,
    "deadlineMisses": null,
    "score": null,
    "outcome": null,
    "terminationReason": null,
    "replayLocation": null,
    "actionableDefects": null
  },
  "limitations": [
    "Source facts are a dated snapshot, not independent reproduction.",
    "Arithmetic is not inference latency or achieved game performance.",
    "The blank record is a proposed checklist, not a runnable evaluator.",
    "No model or game was executed for this documentation audit."
  ]
}