Review methodology
This page describes what every site in the network does with a product category. Each site also publishes its own methodology page covering the cleaning specific to its products.
We do not review products
Nothing here is a review in the usual sense. We do not buy, test or handle the products. The label "review methodology" describes how we handle records — what a manufacturer publishes about a product — not how we handle the product itself.
Gathering the category
A category is gathered by reading listings across the searches that define it, then filtering to the products that actually belong. Search results are full of things that are not the product: parts, accessories, empty enclosures, and neighbouring categories that share a word. Those are removed before anything is counted.
Products are assigned to a class by what the product is, never by the search that surfaced it. This matters more than it sounds. A search for one kind of drive returns dozens of the other kind, and a class built from search terms would be wrong in both directions.
Reading specifications
Every listing is read into a common set of fields for its category, with units converted to one canonical unit per field. A figure is recorded with its source — the structured specification table, or the listing's own free text — because those are not equally reliable and the difference is worth keeping.
Where a listing does not state a field, the field reads not stated. It is never filled with a category default, a typical value or an inference from a similar model. A blank is information; a plausible invention is not.
What gets held back
Three things are withheld rather than published:
- Self-contradicting listings. Where a specification table and the product's own title disagree, the figure is corrected against the model number where that is possible, and excluded where it is not.
- Figures outside physical possibility. Listings sometimes advertise numbers no product of that kind achieves. One such figure is enough to stretch a category's range past every real product in it, so it is treated as the listing failing to state the specification.
- Contradictory evidence. Where a listing's own text supports two different values for a categorical field, neither is recorded. Contradictory evidence is not weak evidence for one side; it is no evidence.
Baselines
A baseline is the median of a field within one class and one size band, with values beyond 1.5× the interquartile range excluded from the calculation so a single outlier cannot move it. Baselines are published only where the class holds at least eight products. Below that a median describes the sample rather than the market, and would move if one more listing appeared — so the page says no baseline can be published rather than printing a number that looks authoritative and is not.
Prices and ratings
We publish price tiers, never exact amounts. A price captured on a scrape date is wrong by the time you read it, and a wrong number presented precisely is worse than an honest range.
We do not publish star ratings or review counts anywhere on any site.
Where this fails
The honest limits, stated rather than buried:
- A manufacturer's published figure can be optimistic or measured under undisclosed conditions. We can check a figure against its own category; we cannot check it against reality.
- Automated cleaning catches the defects it has rules for. Every rule here exists because something got through first — which means the next class of defect is one we have not seen yet.
- Where a listing family covers several variants under one entry, the specifications may describe a different variant than the one pictured.
- Coverage is uneven. Some categories publish a field on nearly every product and some on a third of them, and a field stated by few products supports a much weaker conclusion.
Corrections
If a figure here is wrong, we would like to know precisely which one. A product, a field, and what the listing actually says is enough to act on. Send it here.