Methodology

How we know what we publish.

Each index scores one kind of thing and lives on its own page. There is no combined total.

Index	Measures	Page	Phase
Embodied reasoning	Embodied vision-language models	Models	Phase 1
Robot policy	VLA policies on arms and in simulation	Models	Phase 2
World model	Video and action-conditioned world models	Models	Phase 2
Hardware	Humanoids, hands, joints: rated vs measured	Robots	Phase 2
Deployment maturity	Makers' production, deliveries, commercial traction	Robots	Phase 1
Components	Hands, joints, reducers, screws, motors, sensors, e-skin, arms	Components	Phase 1

Grade	Meaning	Example	Shown
E1 Verified	Verified	We saw it, measured it, or found a checkable record (site visit, our test, public tender award, registry change)	Public
E2 Filed	Filed	A document with legal liability (prospectus, exchange filing, annual report)	Public
E3 Corroborated	Corroborated	At least two independent first-hand sources agree; sources never named	Public conclusion
E5 Company-stated	Company-stated	The company's own claim with no outside support	Public, labeled
Third-party est.	Third-party estimate	An analyst or media estimate; the firm is named	Public, labeled

Data older than 90 days drops one grade or is marked as possibly stale. Two claims from the same source never count as corroboration.

At a 50% success rate, the 95% confidence interval is about:

Every result shows its interval. If we cannot afford enough trials, we do not publish the ranking.

No company can pay to change its rank or to review results before publication.
Paid custom tests never change a company's public score; the method is the same for everyone.
Every commercial or personal relationship is disclosed on the relevant page.
Errors are corrected in public, with a dated log.

Disclosures

Data pending

No corrections yet.