How to read a robotics claim
A practical framework for separating a demonstrated capability from a promise, a projection or a carefully edited clip.
Maintained reference ยท Updated 2026-10-20
Start with the claim, not the footage
Write down exactly what is being asserted: a robot completed one task, completed it repeatedly, worked without intervention, or is ready for a customer environment. Those are different claims and need different evidence. A polished video may establish that an event happened; on its own it says little about frequency, setup, failure rate or what happened outside the frame.
Find the original source
Trace a quotation, number or clip back to its earliest available source: a technical paper, company statement, regulator filing, customer statement or recorded demonstration. Record who made the claim and when. Secondary coverage can point you toward evidence, but should not silently become the evidence itself.
Name the conditions
Useful comparisons require context. Look for the task definition, environment, objects, duration, sample size, software or hardware version, human assistance and failure criteria. If these are missing, say so. Do not infer that an unreported condition was controlled.
Separate observation from inference
An observed event is what a source directly shows or reports. An inference is what that event may suggest. A claim about general capability needs more than one successful example; reliability needs repeated trials and a stated denominator. Keep inference visibly qualified and proportionate to the evidence.
Use uncertainty precisely
Attribute estimates and targets to the speaker and retain their time horizon. Use 'announced target' for a plan, 'reported' for a source's account, and 'demonstrated' only for an observable demonstration. When evidence is unavailable, state that plainly rather than filling the gap with a confident-sounding estimate.