Our Methodology

We don't "feel" that one product is better than another. Every recommendation is reconstructable from documented criteria and linked evidence.

Scoring models

Each product category has a versioned scoring model. A model defines weighted criteria; each criterion maps to specific sourced claims. For the launch robot-vacuum category, criteria include suction power, navigation technology, runtime, noise, base-station functions, mop system, dimensions, warranty, and app requirements.

What we publish — and what stays internal

A deterministic utility score is computed per product and used for sorting and candidate selection. It is an internal ordering tool; we do not present it to readers as a scientific quality rating.
Criterion-level comparisons are published. You see how products compare on each sourced specification, not just a blended number.
Data completeness and research confidence are published separately from performance — for example, "Recommendation confidence: 78% — 7 of 9 important criteria verified."
Public recommendations use qualitative verdicts (recommended, good value, specialist pick) with the underlying facts shown alongside them.

Rules we enforce in code

Unknown is not zero. When a value cannot be sourced, the criterion is marked unknown and excluded from the comparison for all products in the set — it never silently scores as a performance zero. Every criterion value carries a data state: verified, manufacturer claim, calculated, unknown, not applicable, conflicted, or stale.
Every scored claim must have evidence, with primary manufacturer documentation preferred and evidence quality shown explicitly.
Research confidence reflects data completeness and source quality, and is shown next to every comparison — separately from any performance measure.
Model versions are frozen — changing weights creates a new version, so historical orderings stay explainable.

Ownership costs

Total-cost-of-ownership comparisons use documented consumable schedules, energy estimates from stated specifications, and clearly labeled assumptions. We do not track prices over time.

No health claims

We avoid medical and health claims. Products are compared on documented physical, electrical, and functional characteristics.