The RoboScore — How It Works

Robolist.ai exposes two distinct measures. Conflating them would let our data coverage masquerade as a credibility signal — which it isn't.

The RoboScore is published as the number itself. These bands are a plain-English reading of it — they describe a score, they never replace one. The top band is reserved for the most thoroughly-evidenced robots (75+), so it is earned, not handed out.

These bands describe a score; they never replace it. The 0–100 figure is what we publish and what drives the order, so two robots in the same band still rank in a definite sequence. Until 2 August 2026 this table carried bond-style letter grades (AAA … C) — they were retired because a letter grade implies an underwriter, and nobody underwrites this: it is computed nightly from public data.

A category ranking puts one robot against another in the same job — humanoids against humanoids, cobots against cobots — ordered by the RoboScore above, plus a capped news lift explained below. Every category with at least twelve eligible robots gets a board. Robots and companies are never ranked in the same list: a warehouse AMR and the company that builds it are not comparable things, and a list that mixes them is measuring nothing.

Robots first sold before 2020 are held out

A standing that opens with a 2009 research platform does not describe the market a buyer is shopping in today. So a robot we know predates 2020 sits out. A robot with no launch year on record stays in — an unknown year is not evidence of age, and most of the catalog's industrial arms carry no year at all. Treating missing data as “old” would delete real products on no evidence. Each board states how many robots its own gate excluded.

When two robots score the same, we say so

About a quarter of ranked robots share their exact score with at least one other, so the order within a tie matters. Until July 2026 that order was the alphabet, which on some boards handed the top place to whoever's name sorted first. The maker's news lift now separates most of those rows, and where it cannot — two robots from the same maker, or two equally quiet ones — we prefer the newer product, and only then the name. The tie itself stays visible on the page: ordered is not the same as separated, and a medal won inside a tie is a claim the evidence does not fully support.

The week column moves only when the robot's own score moved

Rank position drifts whenever a neighbour moves. Across two archived weeks we measured 3,062 robots changing rank — and 2,591 of them had an identical score both weeks. They had not done anything; the robots around them had. Showing those as arrows would be inventing movement on a page whose entire claim is that it does not. So the weekly change is computed from the robot's own archived score, and a robot that did not move renders as a hold. We also suppress the column entirely across a methodology change — the July 2026 recalibration moved every number, and comparing across it would report our own maths as market news.

News can raise a rank by up to 15 points, and can never lower one

The RoboScore is a slow, evidence-based rating — by design it barely moves, and in a typical week fewer than 4% of robots see it change at all. A weekly ranking built on it alone would be a still photograph. So from August 2026 the boards are ordered by the RoboScore plus the maker's Momentum: a capped reading of recent news — funding, launches, deployments, setbacks — that decays over about a month. That total is the published RoboIndex, so its ceiling is 115: 100 of evidence plus 15 of news.

Three rules keep that honest. Momentum is capped at 15 points, so news can reorder robots whose evidence was already close but can never carry a weak robot past a strong one — we tested an even split of the two scores and it sent a robot rated 19 out of 100, last of 774 industrial arms, to 180th on a single funding round. Momentum is a bonus only: a quiet maker keeps its full RoboScore. And the RoboScore itself never moves for news — it is the evidence verdict alone. That is why a row can sit above a better-evidenced one, and why every row prints both parts of its total.

The 15 is reached at a news volume no maker in the catalogue has yet produced, which is deliberate. We simulated the alternative: setting the cap so that today's loudest maker scored the full 15 flipped the leader on 13 of 25 boards, and on the education board it replaced a robot scoring 63 on evidence, with no news at all, with one scoring 52.8 purely on coverage. At the shipped setting the boards behave exactly as they did under the previous 8-point cap — same number of leaders changed, same worst case — while the scale a reader sees runs to 15 with room left in it.

Momentum belongs to the maker, not the robot — every robot a company owns reads the same news. Left there, a company with eight robots in a category would have its quarter's news counted eight times and its whole catalogue would rise as a bloc; for a spell in August one company held all three humanoid medals on a single press cycle. So the news lands in full on that maker's best-evidenced robot in the category and at half on its others. We simulated the alternatives across every board: counting it in full on all of them left 17 of 28 podiums repeating a maker, half brings that to 12, and giving siblings nothing brings it to 7 — but that last setting goes too far, because those twelve remaining repeats are makers who genuinely do hold the two best-evidenced robots in their category, which is a finding and not a fault. Across those four settings no board's leader changed — though the tie rule below, added later, did move two.

When two of a maker's robots tie on RoboScore, neither is uniquely its best — so neither takes the full lift. Both carry the half, so they end on the same RoboIndex and the board marks them tied, rather than letting a hidden difference in news decide which of two equals ranks higher. That is deliberate. The alternative is a hidden tiebreak handing one of them double the news, which put two robots carrying identical evidence nineteen ranks apart. Where our evidence cannot separate two robots, the board should not pretend otherwise.

A robot board prints that total as its RoboIndex, and prints what the total is made of right beside it: the RoboScore it started from, and the points news added. The two are added, never averaged — we tested an even split and the louder number simply took over the order. Momentum also keeps its own column, because it answers a different question: what is happening now, not what is proven. One more thing worth stating plainly — a maker showing no news is usually our blind spot rather than their silence. Only about a quarter of ranked robots have any news linked to their maker, so we treat “no signal” as unknown and never as a mark against them.

Where a maker has no news signal, that column reads “—”, never zero. A zero would claim the company is quiet; the em dash says only that we have not heard anything, which is the truthful version. It is also a link: if you know about coverage we missed, send us the source and we will check it.

The “new” lens

Some categories also publish a board limited to robots launched last year or later, ranked by the same RoboScore. Newness earns its own list; it never earns a bonus on the main one. A robot with no recorded launch year cannot appear there — the exact inverse of the gate above, and the same rule in both directions: absence of data is never rendered as a verdict.

Each category also has a board of the companies that make its robots. A company's board total is the RoboScore of its best-ranked robot in that category, plus a depth credit of up to 5 points for having more ranked robots there, plus up to 15 Momentum points from news.

The total is a standing within one category, so a company on four boards has four totals. It is not the company's own RoboScore or RoboIndex, which are explained in the next section and shown on each company page.

Buyers pick a brand before they pick a model — the way a car buyer settles on Mercedes or BMW before choosing between an S-Class and a 7-Series. The RoboScore above rates one robot. A maker's RoboScorerates the company, on six factors, rolling its robots' figures up to the brand. Same name, same 0–100 scale, same Momentum added on top to reach a RoboIndex — what differs is only what is being measured, so we always say which. A maker's figure is additive: it never reorders the robot rankings.

Why not all six factors set the number

Activity, Attention and Readiness all depend on how much we have managed to collect about a company. If we averaged all six and simply skipped the missing ones, a company we know less about would score higher — because its weakest factors would be dropped rather than counted. We measured exactly that: a small startup with no recent press outranked one of the four largest industrial robot makers in the world, purely because the big maker had a real Attention score to average in and the startup did not.

Products, Depth and Evidence are present for every company in the catalog, so they carry the rating: 65% for how good the robots are, 5% for catalogue depth, 30% for independent evidence. Activity and Readiness are drawn on the hexagon because they describe the shape of a brand — they cannot move its number. Attention sits between the two: it has no weight of its own, but it is one of the two things Evidence is built from, and it can only ever add. No company is ever rewarded — or punished — for the gaps in our data.

Why catalogue depth is worth five points, not thirty

A company listing thirty robots is not thirty times more credible than one listing a single robot, and the number we can see is partly a measure of how much of a catalogue we have found — our problem, not the maker's. Until September 2026 this rating gave catalogue size 30% of the number, and the result was that size, rather than the robots, decided the order: every one-robot maker started about twenty points behind a twelve-robot one, and no specialist could reach the top forty however good its single robot was. Depth now saturates at eight robots and is worth at most five points — the same cap the per-category company boards have always used.

Missing data is grey, not zero

A hollow grey vertex means we have no data for that factor — not that the company scored badly. Evidence works the same way: every company starts from the same floor, and verified real-world deployments and recent editorial coverage can only ever raise it — neither can subtract, and the floor is identical for everyone, so holding nothing can never be an advantage. We do not hold a missing deployment record against a maker, because the largest robot makers on earth have the fewest such records in our database — a gap in our collection, not in their business.

Attention measures coverage, not quality

Attention counts independent editorial coverage from the last two years, weighted by the outlet. Original reporting counts fully. Syndicated reprints count for very little. Company press releases and newswire distribution count for nothing at all, and five reprints of one announcement count once. A brand being widely written about is a fact about the news, not a verdict on its robots — which is why Attention carries no weight of its own. It adds up to nine points of the hundred from inside Evidence, and it never subtracts: a brand we have read nothing about keeps its full score.

Claiming a company page, verifying it, or paying for any plan changes none of these six factors. A maker who claims their page may submit genuine press coverage we missed, and that goes through the same review as everything else — the article existed whether or not we had indexed it.