Reflexive Co-Embodiment: Synthetic and Public Teleoperation Residual Evidence Package

This dataset record contains the reproducibility and evidence package for "Reflexive Co-Embodiment: Learning Physical Wisdom from Human Intervention Residuals." It includes the 100,000-trial synthetic reflexive co-embodiment benchmark, bootstrap confidence summaries, task/degradation/slice summaries, public teleoperation schema bridges, a small local residual pilot schema, and a DROID/Open X-Embodiment-family public residual benchmark mapping 55,136 action rows from 34 public LeRobot robotics datasets across 422 public episodes into a common human-intervention residual schema. The package supports the paper's claims about intervention residuals, dynamic sovereignty allocation, burden-normalized co-embodied wisdom gain, and source-wise residual normalization. It is intended for reproducibility, reviewer audit, and follow-up analysis. Claim boundary: this evidence package is not a trained-policy leaderboard, not a complete download of the 1TB+ original DROID release, not a complete Open X-Embodiment benchmark, not a real-time human-subject study, and not a physical wearable-robot deployment. Raw residual magnitudes should not be compared across robot embodiments without the included source-wise normalization tables.