Robot vacuums / Roborock / Saros 20

Roborock Saros 20 – AdaptiLift™ Chassis 3.0 for Smooth Threshold Crossing, Powered by 36,000 Pa Suction

Roborock Saros 20 Flagship Robot Vacuum Cleaner exact UK model, compared using published specifications, a timestamped source offer and claim-level evidence limits.

Product score78.7/100Category-specific technical scoreTechnical rank #6
Published suction
36,000 Pa
source ↗ · checked 29 Jul
Battery
5,200 mAh
source ↗ · checked 29 Jul
Navigation
StarSight Autonomous System 2.0
source ↗ · checked 29 Jul
Review verdict#1

Three independent test groups support excellent all-round cleaning, capable mopping, strong obstacle handling and unusually good threshold or low-clearance behaviour. It takes first place, with occasional navigation and mop quirks retained as caveats.

See the full comparison →

The stored offer is older than 48 hours, so price and stock claims are hidden until the next source check.

Retailer listingView retailer listing at Roborock UK Current price withheld because the tracked offer is older than 48 hours
About this result
Secondary specification scopeEligible for specification comparison
Purchase adviceNot currently supported

Secondary specification rank #6 compares published inputs inside the tracked category cohort. It is not the review-led quality result or a reviewed instruction to buy this product.

Candidate-set and method sensitivity are now retained for this leading cohort. They do not satisfy the missing evidence or human review gates.

  • controlled measurement missing
  • claim level trace missing
  • criterion claim ledger missing
  • second person review missing

Reviews checked 30 Jul 2026, 21:39 BST · product data checked 29 Jul 2026, 18:17 BST.

Technical comparisonSpecification context

Specification rank #6 among robot vacuums: published suction 36,000 Pa; battery 5,200 mAh. Published inputs, not our lab test.

Published comparison inputs

Published suction
36,000 Pa
source ↗ · checked 29 Jul
Navigation
StarSight Autonomous System 2.0
source ↗ · checked 29 Jul

Published features

Auto-empty dock

Why the specification score is 78.7

Value at checked price25% weight
100

Score uses the exact model's published specification and checked offer; it is not a laboratory result.

Published suction20% weight
100

Score uses the exact model's published specification and checked offer; it is not a laboratory result.

Navigation and obstacle handling23% weight
62

Unknown binary feature inputs receive the method's neutral half-credit, rather than being asserted present or absent.

Dock and mopping automation20% weight
66

Unknown binary feature inputs receive the method's neutral half-credit, rather than being asserted present or absent.

Evidence coverage12% weight
52

Baseline exact-model source linkage only. Third-party URL reachability does not increase product score.

Evidence report
Product evidence report

What matters, what we know and what is missing

Open evidence report JSON ↗

This is a specification-ranked comparison report, not purchase advice. It uses the category evidence profile below and keeps hands-on findings separate from the published specification score.

Primary real-world questionPickup and coverage

Repeatable cleaning tests by floor, debris, pass count and mapped area.

Hands-on synthesisnot collected

No exact-model hands-on synthesis is published. This report covers sourced comparison inputs, coverage and open gaps; it makes no claim about real-world performance beyond retained sources.

Specification rank#6

Inside the bounded published category method.

Stress-test range#1–#15

Across 15 factor-removal and weight-change scenarios.

Purchase adviceBlocked

4 blocking gaps remain.

Official sourceRetained
Independent editorial1
Owner pools1
Retailer observations1/2
Controlled measurements0
Human second reviewNot completed
Category evidence needs, decision cruxes and constraints
Pickup and coverage

Repeatable cleaning tests by floor, debris, pass count and mapped area.

Hands-on state: not collected
Obstacle avoidance and navigation

Transparent tests with named obstacles, lighting, layout and repeat runs.

Hands-on state: not collected
Mopping

Separate stain, edge, pressure and carpet-avoidance evidence.

Hands-on state: not collected
Dock automation

Repeated evidence on emptying, washing, drying, refill and fault handling.

Hands-on state: not collected
Thresholds and edges

Measured thresholds and direct edge or corner cleaning observations.

Hands-on state: not collected
App and privacy

Hands-on feature and reliability evidence, plus sourced data-handling claims.

Hands-on state: not collected
Maintenance and reliability

Repeated-use evidence on tangles, consumables, cleaning burden and failures.

Hands-on state: not collected

Hands-on gate: Resolve the exact robot, dock and region. Require transparent cleaning and navigation tests; mopping and dock conclusions need their own evidence.

Decision cruxes
Obstacle and navigation performance

Published navigation features do not establish repeatable avoidance, coverage or recovery performance.

Pickup and mopping performance

Published suction does not establish pickup, stain removal or dock maintenance under shared conditions.

Buying constraints
Warranty
2 yearssource ↗
Returns
30-day direct return policysource ↗
Safety
Use grounded power and the approved battery and charger; the manual also identifies indoor-use, hot-water, hot-air and wet-floor risks.source ↗
Installation or setup
not structured
Normalized decision receipt

7 mapped fields each carry a canonical receipt hash and an origin status. Source access time is kept separate from field observation, so a configured value is not presented as freshly re-extracted.

sha256:1807a9ea3b8d7aa15c75b4f704b05aaac8436543cdeda12ce4353c094fc4af7d

The score and rank compare retained specification inputs inside this bounded category cohort. They are not a hands-on quality verdict and are not comparable across categories. Open evidence gaps block purchase advice regardless of specification score, rank, offer state or review-synthesis verdict.

Source notes and limitations

Evidence limitations

  • Published specifications are not Agent Review Data laboratory measurements.
  • The manufacturer-direct source offer and availability can change after the checked time.
  • Exact model identity is keyed to the model/SKU in the source register; related colours and bundles are not merged.
  • Official CE English user manual was checked 2026-07-29 as a supporting manufacturer source.
  • The 8.8 cm claim applies to double-layer threshold handling under stated product conditions; it is not reduced to an unconditional millimetre field or scored.

Recommendation caveats

  • The score compares published specifications and a checked offer; it is not a controlled performance test.
  • Scores are comparable only within this category.
  • Published specification inputs can use different test conditions; the retained values are not treated as laboratory-equivalent.
  • Separate third-party sources are shown as contextual evidence, not as a universal consensus.
  • The 8.8 cm claim applies to double-layer threshold handling under stated product conditions; it is not reduced to an unconditional millimetre field or scored.

Owner feedback across the web

One exact-model manufacturer-hosted owner pool is visible. Only its dated count is retained, so no rating synthesis is calculated.

Roborock UK product page/5 · 21 reviews

The exact manufacturer product page displayed 21 reviews; the native aggregate rating was not retained.

Inspect review pool ↗

Independent tests and owner discussion

Independent exact-model testing adds cleaning, obstacle-avoidance and threshold context to the official specification and manual.

Expert Reviewseditorial test · exact model · included

Roborock Saros 20 review: Hands-on exact-model testing of cleaning, object avoidance and threshold handling.

Checked 29 Jul 2026, 01:00 BST · Independent editorial context; the published 36,000 Pa claim remains manufacturer evidence.Inspect source ↗
  • Third-party evidence is claim-level context and is not a substitute for a controlled Agent Review Data test.
  • Reddit is a self-selected owner discussion source, not a representative survey.
Claim-level trace not yet modelled · 2 of 160 available

This record exposes source-linked inputs and their provenance status, but it does not yet model individual claims, relations and criterion contributions.

Inspect the trace build-out →

Sources