Most sites describe their standards in prose. Ours are enforced by software: a page on this site literally cannot be published if it breaks the rules below, because the build that produces the site checks them and fails. This page explains the rules in plain language.
Every substantive claim on this site is a structured record with a version number, the evidence it cites, a confidence level, a verification date and a review deadline. A claim that cites nothing is an error, not an opinion. When a claim passes its review deadline, presenting it as current becomes a build failure — that is why our pages say when they were last reviewed.
Sources are stored immutably and graded. Material from a party with a commercial interest in the conclusion is treated as a claim to be checked, never as evidence in its own right.
When two studies disagree, the disagreement is recorded on the claim itself and must be visible in the published text. We are not allowed to present a contested finding as settled — mechanically not allowed: the build rejects it. You can see a live example on the toilet-health evidence review, where a 2019 and a 2022 study disagree and we say so.
Any claim can be challenged by an audit that pre-registers, in advance, what evidence would prove it wrong. A delivered audit must be answered on the record: accepted, or rejected with a written reason of substance. Silence is not an option — an unanswered audit blocks the build. An audit that produced no new sources cannot raise a claim's confidence.
When we get something wrong, we change the page and say what changed on the changes page — including recommendations we withdrew because the arithmetic stopped supporting them, and what that cost us in commission.
Product prices change. Where a price appears on this site it carries the date we last verified it against the seller's page; a price past its review date is treated as stale and re-verified before it may be shown as current.
Affiliate commissions (declared on how we make money) live downstream of the calculators. The recommendation has to survive the arithmetic first; we have dropped better-paying products when the numbers pointed elsewhere, on the record.
Infallible. It makes us checkable — which we think is worth more.