Corrections and score changes

Exact-model specifications and buying guidance corrected

Shark Matrix RV2310AE is a vacuum-only model. Booster T2 editions and AGIBOT configurations now use their own manufacturer specifications. Apollo 2 is an enterprise platform unveiled June 30, 2026; its old public price reference was unsupported. NEO ownership, subscription and deposit terms are distinguished. Saros navigation and mop designs are separated by model, and the Roborock/Dreame comparison no longer assigns performance winners without a comparative evidence package.

October availability observations supplement the preserved September history. New Saros 20 Flow and US Navimow X4 records remain unscored. Existing RAI scores have not been recalculated from unverified claims. Older article prices remain dated snapshots.

Roborock S8 MaxV Ultra generation mix-up corrected

Replaced “ReactiveAI 3.0” and “StarSight 2.0” references with PreciSense LiDAR and Reactive AI 2.0; replaced VibraRise 2.0/3,000-per-minute copy with VibraRise 3.0/up to 4,000 per minute; and corrected the claim that the model has no edge cleaning by identifying its FlexiArm side brush and edge-mopping system.

RAI v1.0 established as the canonical formula

RAI is now explicitly versioned as the rounded equal-weight mean of four autonomy pillars. Category performance such as cleaning, mopping, pet hair, and value is separate. Conflicting five-factor copy was retired from publication.

Three conflicting composites normalized to RAI v1

The displayed composites are now calculated directly from their four published pillars: Husqvarna Automower 415X changes from 84 to 85, iRobot Roomba Max 705 Combo from 92 to 93, and Narwal Freo X Ultra from 89 to 90. The upstream conflicts remain visible in the data-quality queue until their source records are migrated.