Editorial analysis
A taxonomy is different from a failure log
Five records name errors retained in the AGIBOT World dataset, including missed grasps and dropped objects. Their counts are unrecorded. The existence of a category does not tell readers how frequently it occurred.
Zero observed events needs an exposure measure
The other records describe manufacturer validation claims. The denominators use different exposure measures and should not be combined. Zero reported events within a stated observation window does not establish a zero probability of future failure.
What a useful failure comparison requires
A comparison needs a consistent event definition, observation period, workload and intervention policy. Severity and event count should remain separate. Until a comparable log is available, this page remains an evidence note rather than a Pareto ranking.