What counts as evidence
A claim about a product on this site comes from one of three places, and which one is always stated:
- A test we ran. Reported with the date it was run and a link to the protocol.
- A vendor's own documentation. Reported as a vendor claim, with the date it was checked. A capability we did not exercise is never described as a test result.
- A third-party source. Named, dated and linked, and distinguished from primary material.
Anything outside the protocol stays unscored and is named as untested. We would rather print a gap than imply a number covers something it does not.
Handling our own product
Slaide is built by the team that publishes this site. The safeguards are structural rather than promissory:
- The rubric and its weights are published before results, and are identical for every product.
- The overall score is computed from the six subscores. There is no field in which a flattering total could be stored.
- Slaide's limitations appear on every page that recommends it.
- Category awards go to whichever product earned them; Slaide holds one, not five.
- Every page that ranks, compares or reviews products carries the disclosure.
Commercial independence
No vendor can pay for a position, an award, a review or a mention. There are no affiliate arrangements with competing products, and no advertising. Links to Slaide are marked rel="sponsored"; links to competitors are unpaid and carry no tracking.
Use of AI in producing this site
AI is used for research, drafting, structuring and code, which is unremarkable and would be true of most publications now. It is not used to invent results. Every figure on this site traces to a test that was run or a source that was checked, and the distinction between those two is recorded in the data rather than left to prose.
When a score changes
A score is revisited when a vendor materially changes the capability being measured, when a category benchmark is re-run, or when a reader demonstrates that a result does not reproduce. Changes are dated and the reason recorded.
Pages carry three separate dates — published, last updated and last tested — because they mean different things. We do not restamp a page as fresh because it was rebuilt.
Corrections
Corrections are published, dated and kept. A research page carries a changelog; an article carries a correction note. We do not silently overwrite a published result, including when the correction is unflattering to us.
If a result does not reproduce, or a vendor believes a test misrepresents their product, tell us what you ran and what you saw. A reproducible counter-example changes the page.
What we do not do
- Publish a page for every phrasing of a query when one page answers all of them.
- Score a product on a capability we did not exercise.
- Mark up our editorial scores as review ratings for search-result stars.
- Restamp publication dates to look current.
- Present a vendor's claim as our measurement.