How We Pick
The complete method: five weighted metrics, a worked example you can check, and an explicit account of what we do not do.
Every score on this site comes from five metrics with published weights: suction 25%, navigation 25%, pet hair 20%, running cost 20%, setup 10%. Each metric is scored 0-10 from the manufacturer's own published specification. The arithmetic is on this page, so you can recompute any score on the site and disagree with a specific step rather than with the conclusion.
The claim we refuse to make
Every credible competitor in this category ranks on the strength of a test lab. Vacuum Wars states it has independently tested over two hundred machines. RTINGS runs a documented protocol in a real room. TechRadar's top-ranking page is literally titled around having tested at every price point. Those are genuine credentials and they earn the rankings that come out of them.
We have no lab, and we are not going to pretend otherwise. Everyone in this category says they tested twenty machines. We have not tested any, and we say so - on this page, in the byline of every article, and in an explicit section on every single review naming the specific claims we could not check.
What we do instead is the thing almost nobody in this field bothers with: compile the published specifications, read what the manufacturers actually document, do the arithmetic, and publish all of it so you can check the work. When we surveyed the pages ranking for these searches, not one published a reproducible scoring formula. That gap is the whole reason this site exists.
The rubric
| Metric | Weight | What it scores | Where the figure comes from |
|---|---|---|---|
| Suction | 25% | Published Pascal rating, and whether one is published at all | Manufacturer listing. No figure published scores as unverified, not as good |
| Navigation | 25% | Sensor type: bump, gyroscope, LiDAR, dToF, structured light | Manufacturer listing. 'Smart navigation' with no named sensor counts as no map |
| Pet hair | 20% | Brush geometry: brushless, anti-tangle, self-cleaning, and whether the side brush is covered | Manufacturer listing. Suction is deliberately not credited here |
| Running cost | 20% | Dock consumables over three years, plus filters and brushes | Published dock intervals, converted arithmetically to unit counts |
| Setup | 10% | How much fighting the machine takes to start working | Published control methods, Wi-Fi requirements, app maturity |
A worked example, so you can check the arithmetic
Take the Ecovacs DEEBOT N20 Plus. Its five metric scores are suction 8.5, navigation 7.5, pet hair 8.0, running cost 9.0, setup 7.5. Applying the weights:
- Suction: 8.5 x 0.25 = 2.125
- Navigation: 7.5 x 0.25 = 1.875
- Pet hair: 8.0 x 0.20 = 1.600
- Running cost: 9.0 x 0.20 = 1.800
- Setup: 7.5 x 0.10 = 0.750
- Total: 8.15, displayed as 8.2 out of 10
Every score on this site is computed this way, and every metric score is shown on the product card so you can see exactly which one you disagree with. If you think running cost should carry more than 20% because you hate buying dust bags, raise it and rerun the numbers - the ranking will change, and that is the correct outcome rather than a flaw.
How each metric is actually judged
Suction, 25%
We score the published Pascal rating, with a heavy caveat: there is no shared test standard between manufacturers, so two brands' figures are two different measurements sharing a unit. We treat large gaps as meaningful and small ones as noise.
A machine with no published figure scores as unverified, not as good. Three well-known brands here - iRobot, Shark, and eufy on one listing - publish no Pascal rating at all, and it costs them points. That is deliberate: on a site that ranks by published specification, silence has to have a consequence, or the ranking rewards non-disclosure.
Navigation, 25%
Scored by sensor type, because that is what determines capability. Bump-and-infrared scores low; gyroscope-assisted patterns slightly higher; LiDAR substantially higher; dToF a little higher again; structured-light obstacle avoidance highest of all, because it is the sensor that decides whether a machine finishes a run on a real floor.
A listing claiming 'intelligent path planning' or 'smart navigation' without naming a sensor is scored as having no map, because in every case we checked, that is what the phrasing described.
Pet hair, 20%
Scored on brush geometry, not suction. Hair is light and almost any machine lifts it; the failure mode is hair wrapping around the brushroll until it stops turning. So we credit a brushless inlet, a named anti-tangle geometry, or a self-cleaning brushroll - and we credit an anti-tangle side brush most of all, because it is the worst wrap point and almost every manufacturer ignores it.
Running cost, 20%
Derived arithmetically from published dock intervals over three years of daily running. A bagless dock scores highest because it has no consumable at all; a 90-day bag scores above a 60-day one; a machine with no dock scores well on cost and is penalized elsewhere for the bin-emptying it demands.
We do not quote consumable prices, because they vary by retailer and by whether you accept generics, and a price written into a sentence is stale the next day. We publish the unit counts and let you multiply by whatever the parts cost you.
Setup, 10%
The smallest weight, for the thing that matters most in week one and least thereafter. It covers control methods, whether a physical remote exists, the 2.4 GHz Wi-Fi requirement almost every machine has, and how mature the app platform is.
What we do not do
- We do not claim to have tested anything. No 'in our testing', no unit counts, no lab.
- We do not emit review-rating structured data. The score on the page is editorial content and it stays there. Marking it up as a review rating would tell search engines we tested these machines, in a format only they read.
- We do not write a price into prose. Prices render only from the live data layer with a verification date, or the button says 'Check price'.
- We do not paraphrase owner reviews into our own voice. Where an owner complaint is worth reporting, it is attributed and linked.
- We do not publish testimonials. This brand is new and has no customers to quote; inventing some is the fastest way to deserve none.
- We do not recommend a machine that is out of stock. Availability is checked against live retailer data before a page is written, and several models we had planned to cover were dropped for exactly this reason.
Where the figures come from
Every specification on this site is read from the manufacturer's own product listing or documentation, and the specific document is cited on every page with the date it was read. Where a manufacturer does not publish a figure, the page says 'Not published' rather than estimating one or borrowing a number from a similar model.
Prices and stock status come from a live data layer that is re-checked daily and stamped with the date it was last verified. If that data goes stale beyond 48 hours, every price on the site disappears automatically and the buy controls fall back to 'Check price' - which is the honest output, not a degraded one.
Corrections
If a figure on this site is wrong, we want to know, and fixing it takes priority over everything else. Tell us here and cite where the correct figure is published. Our full correction and update policy is on the editorial policy page.
Questions people actually ask
Have you tested these robot vacuums?
No. Everything on this site comes from published specifications, manufacturer documentation and retailer listings, all cited with the date they were read. Every review carries an explicit section naming the claims we could not verify.
Why should I trust a ranking from someone who has not tested the machines?
Because you can check it. The five metrics, their weights, the worked arithmetic and the source for every figure are all published, so you can recompute any score and disagree with a specific step. A test-lab ranking asks you to take the result on faith; this one hands you the working.
How do you score a machine that does not publish a suction figure?
As unverified, not as good. Three well-known brands here publish no Pascal rating at all, and it costs them points. On a site that ranks by published specification, silence has to have a consequence or the ranking rewards non-disclosure.
Can I change the weights?
That is exactly why they are published. If running cost matters more to you than 20%, raise it and rerun the arithmetic with the per-metric scores shown on every product card. The ranking will change, and that is the correct outcome.
Why do your pages not show star ratings in Google?
Because review-rating markup is a machine-readable claim that this publisher tested and rated the product. We did not, so we do not say it - not in prose and not in a format only a search engine reads. It costs us the star ratings competitors show, and that is the correct trade.
Sources
Every specification quoted above was read off one of these documents on the date shown. Where a manufacturer does not publish a figure, this site says "Not published" rather than estimating it.
- Manufacturer listings for every machine on this siteEach page cites the specific listing its figures were read from, with the date.
- Live price layerPrices and stock status re-checked daily and stamped with a verification date. Stale data disappears automatically after 48 hours.