This field publishes figures spanning four orders of magnitude, and the spread is mostly not disagreement about physics. It comes from two things that are easy to compress into a single number. The first is what was measured: a figure can come from simulating a modelled device, from an emulation on reconfigurable hardware, from a projection to the final form of a technology, or from silicon running a workload. All four are ordinary engineering practice and each answers a different question. The second is the denominator. Energy per operation describes a device. Energy per completed task on a body includes actuation and is bounded by it. Time per action is a latency against a control period. A figure of seventy thousand in the first denominator and a limit of two point four four in the second are both correct, because they measure different quantities. Sorting once on these two axes tells you more than reading any number of headline claims, because a pattern appears: figures obtained on silicon running a robot workload sit in single digits, and figures above a thousand come from simulation, emulation, projection, or a device-level measurement in the per-operation denominator.