The short version
- We buy every unit at retail. Where a manufacturer lends one, the review says so on its face.
- Testers do not see the price of a product until scoring is finished.
- Every score is the weighted average of a published breakdown, computed rather than typed.
- Each category has a minimum test period, and reviews are not published early.
- When we are wrong we correct the page, date the correction, and describe it.
Every review site says it is rigorous. The useful version of that claim is a list of things you could check, so this page is a list of things you could check.
Where the products come from
We buy them, at retail, at the price you would pay. In the last twelve months that covers 218 of the 224 products we have tested.
The remaining six were loan units from manufacturers, all for products that had not gone on sale. Every one of those reviews says so in the panel at the top, in the same line that gives the test duration. We do not accept gifted units, and we do not accept units on the condition that we return them by a date.
The price is hidden until the end
Testers work from a spreadsheet in which the price column is blank. It is filled in when the subscores are locked.
This is not a moral achievement — it is a mitigation for a specific, well-documented effect, which is that knowing something costs $900 makes people find reasons it is good. It does not survive contact with an obviously premium product, and we do not pretend otherwise. It does mean that “for the money” reasoning enters the process once, deliberately, at the end.
Minimum test periods
A review is not published before its category’s minimum, whatever the deadline.
| Category | Minimum |
|---|---|
| Audio | Four weeks of daily use |
| Computing | Four weeks as a primary machine |
| Cameras | 2,000 frames |
| Kitchen | 200 preparations |
| Fitness | 400 miles, or six weeks for wearables |
| Workspace | Six weeks of working days |
These are floors, not targets. The Stride 5 review took eleven weeks because 400 miles takes eleven weeks.
The score is computed, not chosen
Every review carries a breakdown of weighted subscores, and the headline number is the weighted average of them. It is calculated by the site at build time from the numbers in the article’s own frontmatter.
This is a deliberate constraint on ourselves. An editor cannot decide a product “feels like an 8.5” and then arrange the subscores to suit, because the badge and the table are the same data. If you disagree with our weighting — and the weights are published on every review — you can do the arithmetic yourself.
The bands
Scores land in one of five bands, and the band is what we would actually say out loud. A 7.2 is not “good” — it is “good, with caveats”, and the caveats are the reason the article exists.
We do not use the top of the scale often. In two years, three products have scored above 9.5, and we expect that to stay rare, because a scale where everything good scores 9 is a scale with three usable values.
Corrections and re-tests
Guides carry a revision date and a note describing what changed. Reviews are re-opened when firmware, pricing, or a competitor moves enough to change the answer.
If we get something wrong, we correct the page rather than issuing a separate one, date the correction, and describe what was wrong. Withdrawn recommendations are stated as withdrawn. A page that quietly stops mentioning a product it used to recommend is a page that is lying by omission.
What we do not do
Sponsored reviews. None, at any price, in any format.
Pre-release embargoed scores. We will not agree to publish on a date set by a manufacturer, because that date is always before the minimum test period.
Review copies of software or subscriptions. We pay, on the same tier a reader would.
Manufacturer-supplied measurements. We publish our own, and where the two disagree we say so. Where they match — as Northbeam’s did to within 1.2 dB — we say that too, because it is the more useful information.
