{
  "passed": true,
  "excluded_from_evaluation": true,
  "recommendation": "Resume with at most two simulator workers and unchanged per-run audits; parent owns launch.",
  "raw_evidence_remote": "/home/parclab/yijia/hybrid2/work/g1_factory/concurrency_v3/runs/20261112T220102Z",
  "runner_sha256": "ce4ff8d16c4b11b85b4628799de59ce352d1ca3da7cf9073b70c36b6437c6da8",
  "runtime_freeze_sha256": "1b63b1d5495fd24fa8d9b558c81e4be971f3d7154d6ec769155bee5fa79a4658",
  "validation_sha256": "8214b7a9fba1f5c12bac79b72d021a4f6722972a3495ebee2b7d139a00646e7f",
  "supplemental_validation_sha256": "8e624357f086c514bf92c6b514e91c70ca51543913b5877cf7eb2b9ff23cce8a",
  "manual_review": {
    "completed": true,
    "method": "Viewed all four 5x5 contact sheets, one for each task and view, derived from the 25 original packets at steps0,40,...,960.",
    "images_sha256": {
      "review_images/LMDrillLift_overview_rgb.png": "5c45ab39dbc36a08d97a102b20aca9d79a88442427e0d4eafc2bb778f6e7e42b",
      "review_images/LMDrillLift_rgb.png": "a900fc7c23c8623b06cf436231b6f14d3d47897e30ddcb2570ee20a833a4ca70",
      "review_images/LMPushShelfForward_overview_rgb.png": "76470f6825b187671ddf318f0cf71dd8706027782b3e8e3884c0c96e718acc68",
      "review_images/LMPushShelfForward_rgb.png": "ebeb5b90a4264ab00ba69959082a59316a55fb0d2b9d18c82e85dbcd049266c3"
    },
    "observations": [
      "All 100 sampled views contain visible scene geometry; no black, empty, or visibly corrupted sampled frames.",
      "Drill head view shows the robot hands, floor, table and drill. Overview shows the stationary standing robot, table and drill.",
      "Shelf head view shows the robot hands, shelf and warehouse. Overview shows stationary standing robot, shelf and target.",
      "Initial exposure is darker than later samples but scene geometry is clearly visible. No claims of bitwise photometric equivalence are made."
    ],
    "coverage": "Global mean/std/min/max checks cover1000 frames per view per task. Exact spatial variance/percentile checks and manual review cover25 raw packets per view per task; intervening images are not individually reviewed.",
    "supersedes_pending_flag_in": "supplemental_validation.json"
  },
  "summary": {
    "shared_control_window_seconds": 252.38803887367249,
    "observed_aggregate_control_steps_per_second_in_shared_window": 7.849922098996894,
    "pair_process_makespan_seconds": 277.64722084999084,
    "peak_rtx_memory_mib": 6392
  },
  "checker_revision": {
    "original_sha256": "bba7f62387638329e72a29f9f2942153a69857773ba05ce6cfef63fe39026e4d",
    "revised_sha256": "3a774d50bb6ce3ad022a82ffc80169b369c3311d63589dc900b8d589f5718970",
    "reason": "Original exact float-dictionary equality rejected np.std recomputation after strided live pixels were serialized into contiguous saved arrays. Observed max difference2.842170943040401e-14. Revised read-only checker permits mean/std absolute difference<=1e-12; shape,dtype,min,max and raw packet hashes remain exact. No simulator, observation, model, protocol or frozen source changed.",
    "original_source_preserved": true
  },
  "limitations": [
    "Standing controls do not retest learned-model RPC/session interleaving.",
    "Image statistics and physics/FK checks are necessary evidence, not proof of pixel-perfect or bitwise dynamic equivalence.",
    "Observed throughput includes extra read-only logging; startup and production audits limit extrapolation.",
    "Task success is not a pass criterion and no diagnostic enters the evaluation dataset."
  ]
}