repro-bundle/outputs/summary.json

{
  "bigym_sup": {
    "success": 0.67,
    "speedup": 2.01,
    "tasks": 20,
    "trials_per_task": 100
  },
  "libero_pi05_sup": {
    "success": 0.973,
    "speedup": 1.35
  },
  "libero_vla_adapter_sup": {
    "success": 0.937,
    "speedup": 1.34,
    "qualification": "Different base policy; not a controlled same-policy comparison."
  },
  "real_world_sup_mean_success": 0.6111111111111112,
  "real_world_demo_mean_success": 0.6,
  "real_world_sup_speedup": 2.17,
  "real_world_demo_speedup": 2.07,
  "scheduler_parameters": 5120000,
  "demospeedup_parameters": 4000000000,
  "parameter_ratio": 781.25,
  "evidence_qualifications": [
    "Parameter count is in Table 3, not Table 4.",
    "ADM-vs-MLP evidence is Figure 7, not Table 5.",
    "No official SuP implementation or robotics checkpoints were found; values are paper-table audits, not reruns."
  ]
}