How the score works.

Five pillars, one hundred points

Every robot in the catalogue gets the same treatment. The pillars are weighted for a buyer, not for an investor: what the thing does, what it costs, whether you can get one, what surrounds it, and what it has actually proved.

Value is judged inside a price bracket

A $299 desktop robot is compared with other $299 robots, not with a $70,000 research platform. Otherwise the cheapest item on any list always wins on value, which is useless to somebody deciding how to spend $15,000. A machine scores well on value when it offers more than its immediate neighbours in price.

Buyability is a real pillar, not a footnote

Most humanoid coverage treats availability as a detail. For a buyer it is the whole question. A machine with a published list price on the maker's own store, in stock, shipping worldwide, scores the full twenty. A machine with an impressive demo, a press contact and no order page scores close to nothing, no matter how good it is.

Proof is where almost everything scores badly

Ten points for units in the field, named customers and a company that looks like it will still exist in three years. Most humanoids score one or two. That is not a flaw in the scoring, it is the state of the industry, and any ranking that hides it is selling something.

Where we overrule the arithmetic

A computed score cannot know that a cheap machine will not run your own code, or that a company has quietly stopped answering email. When we adjust a pillar by hand, the robot's page says computed, then adjusted by hand instead of computed from the data. Right now 2 of 44 orderable machines carry an adjustment.

Provisional scores

We hold twelve core specification fields per robot. When fewer than half are known, the score is marked provisional on the page, because a score built on four facts is not the same claim as one built on twelve. That currently applies to 24 orderable machines.

How prices are graded

Not all prices are equal, and pretending otherwise is how buying guides go wrong. Every price on this site carries the level of evidence behind it.

• List price from the maker. Taken from the manufacturer's own store or product page.
• Street price, reported. From trade press or a reseller listing. Configurations vary, so treat it as a range.
• Third-party estimate. Nobody publishes a price. Somebody guessed, and we say who.
• No published price. A sales conversation, a lease, or nothing at all.

Independence

We take no commission on any sale, run no affiliate links, and no maker pays to appear or to rank. Links to sellers are marked nofollow. If that ever changes, it will be written here first and on every page it touches.

Corrections

Prices move and we will sometimes be behind. Every price carries the date we checked it. Send a source to hello@humanoid.observer and a sourced correction goes live within a day.