{
 "format": "search_record_v2",
 "cameras": [
  "countertop_camera",
  "right_camera",
  "left_camera"
 ],
 "chunk": 50,
 "state": "named object poses, articulation root/qpos/qvel, the 14-dim robot command, and one JPEG per camera \u2014 all at the node, captured where the search stood"
}