Independence and disclosure
A review site is worth reading only if you know what moves its verdicts. This page says so plainly, and commits us before there is any money to argue about.
Draft — not yet reviewed by a lawyer
This page was drafted from how the platform actually works. It has not been reviewed by a qualified lawyer and is not legal advice. BRD 14.3 makes that review a launch dependency. Set NEXT_PUBLIC_LEGAL_REVIEWED=true once it is done, and fill in the operator details.
As things stand
- Nothing on this site is paid for. No brand, retailer, agency or intermediary has paid, or been asked to pay, for coverage, placement, a verdict, or the timing of one.
- There is no advertising and no affiliate link on any page.
- We hold no position in any company whose products are reviewed, and accept no free products, loans or hospitality.
- Subjects are chosen by search demand and recency, not by who is willing to be covered.
If that changes
Affiliate links, data licensing and category sponsorship are all possible later. These rules apply from the moment any of them starts, and they are the terms on which we would take the money:
- No commercial relationship reaches the debate. The agents receive the evidence base and nothing else; there is no field in which a commercial arrangement could be expressed.
- Affiliate links, if used, are applied identically to winners and losers. A verdict that says avoid this and buy that must earn the same commission either way, so there is nothing to gain by changing it.
- Any commercial relationship is disclosed on every page it touches, at the top, not in a footer.
- Sponsorship, if taken, is of a category hub and is clearly labelled. A sponsor never learns a verdict before publication and never influences one.
- We will not accept payment to remove, delay or soften a verdict, and we will say so publicly if it is offered.
AI, disclosed
The arguments and verdicts here are generated by large language models and reviewed by a person before publication. We do not present them as human-written. Every page names the model that argued each role and the version of the instructions it was given, and the methodology page reproduces those instructions in full.
The agents may not argue from memory. Each is restricted to a curated evidence base, and a claim that cites nothing is discarded before publication rather than softened.
The Judge deliberately runs on a different provider from the two arguing agents, so one model’s habits cannot decide outcomes across the site.
Where the evidence comes from
Manufacturer specifications, dated street-price observations, independent benchmarks, professional reviews (linked and paraphrased, never reproduced), aggregated complaint patterns, service and repair reports, software update histories, and evidence submitted by readers and by brands. Every item carries its source, the date it was observed, and a confidence rating, and all of it is listed on the page it informs.