How this library is made

An article is written only from the full text of the paper it discusses. Abstracts are used to decide whether a topic is worth taking on, and for nothing else — they omit the sample size, the dropouts, the arm that did not work and the funding.

Where a paper cannot legally be read in full, one of two things happens: the topic is dropped, or the article says in the text that the paper was not read and why. Paywalls are not circumvented.

Each DOI is checked against Crossref and must resolve to the paper actually being cited — the right authors, the right journal, the right year. No identifier is written from memory. An invented DOI is worse than no citation at all, because it looks checkable and is not.

Before an article built on a review goes out, the papers citing that review are checked for newer or contradicting work. A larger or more recent study that disagrees has to appear in the text. Leaving it out would be misleading whether or not any rule required otherwise.

A large trial that failed to find an effect for an ingredient we sell belongs in the article about that ingredient. This is the rule most easily broken by accident and the one we check hardest, because the commercial incentive runs against it.

Conflicts of interest in the studies we cite are disclosed in the text — manufacturer funding, manufacturer co-authors — and so are the weaknesses: twenty participants is written as twenty participants, and a study in mice is written as a study in mice.

Each piece carries one of three labels, chosen from the sources it actually stands on rather than from the subject:

• Strong evidence — Meta-analyses or randomised trials in humans, pointing the same way.
• Moderate evidence — Human studies exist, but are limited in size, population or consistency.
• Preliminary evidence — Mechanism or early findings only — largely animal, cell or unpublished work.

The researchers whose work we discuss are named. A person enters our register only where a genuine public professional profile was found and confirmed — never guessed from a name. Each profile links back to the articles that discuss their work, so a reader can see exactly what we claim they found.

Health claims follow Regulation (EC) 1924/2006 and the authorised list in Regulation (EU) 432/2012. A claim is quoted in the register's own wording, attached to the specific nutrient it belongs to, with the percentage of the nutrient reference value supplied — never attributed to a product as a whole.

For ingredients with no authorised claim, we say so plainly rather than describing a mechanism in a way that implies one. Titles are written as questions or as descriptions of the state of knowledge, never as promises — a title is commercial communication and is bound by the same rules as the text.

Cosmetic products in the range are governed by Regulation (EC) 1223/2009 and carry no health claims at all.

When an article is wrong, the article itself says so. The correction is dated, the original claim is left visible, and where the mistake came from a failure of method, the article explains the failure. We would rather be publicly wrong once than quietly wrong permanently.