About

How these pages are made

Every page on this site is built from a record: a list of studies retrieved from public indexes, quoted sentence by sentence, labelled by the kind of evidence each is, and checked by a named person before it is published. This page describes that process and its limits.

The method, in order

1. Retrieve. For each compound, every publication in Europe PMC whose title or abstract names it or a listed alias is fetched, with its MEDLINE publication type and subject headings. Compound identity, clinical phase and approval status come from ChEMBL. Counts on a page are counts from this retrieval, dated.
2. Classify. Each paper is labelled by study design and species from its own indexing, with a text fallback for papers not yet indexed. Reviews are kept as background reading and never become claims.
3. Quote, do not paraphrase. A claim on this site is a sentence copied verbatim from a paper's abstract, with the study's design, sample size, species and year in front of it and the source identifier after it. Nothing is combined across papers. Where a dose appears in a quotation, it is attributed to the compound only when its name sits beside it; comparator doses in the same sentence are left out.
4. Suppress, do not fill. A section with no evidence is not rendered. A compound with no indexed studies gets a short page that says so. A zero is reported as a zero.
5. Fill the gaps, labelled. Where the literature is silent on something a reader needs, mechanism, development history, regulatory status, or what is commonly reported outside the studies, the page says it from general knowledge and marks the passage editorial synthesis, so a cited finding and an editorial passage are never confused.
6. Review, when it happens, is shown. When a named person with a stated credential has checked a record, their name, credential and date appear at the top of the page. Pages are published as compiled and improved in place; every change is dated in the changelog.

Evidence tiers

Every claim carries one of six labels. They describe the kind of evidence, not its conclusion: a compound with one human trial and forty animal studies is described by both facts, not the better one.

Approved label

A regulator has reviewed the evidence and approved a label. The strongest tier, and still specific to the approved use and population.

Human clinical trial

At least one indexed clinical trial in people. The trial's size, design and outcome are what matter, and the evidence table shows them.

Observational, human

Human data without randomisation: cohorts, case series, registries. Association, not cause.

Animal, preclinical

Studies in animals only. Findings do not transfer to people, and weight-normalised doses do not scale between species.

Mechanistic, in vitro

Cell or tissue work. Shows what a molecule can do to a receptor or pathway, not what it does in a body.

Community-reported

No indexed study. What circulates is reported as circulating, sourced to where it appears, and never restated as fact.

Editorial synthesis

Written from general knowledge rather than a ledger source: mechanism, history, class context, regulatory status, common reports. Marked as editorial synthesis wherever it appears.

Publication and review

This site is in its pilot. It holds 34 compound records, 8 stack records, 10 comparison records and 3 tools, built on 1,016 sources and 1,121 quoted claims.

Pages are published as they are compiled and improved in place. 28 of 34 compound records carry a named reviewer's sign-off, shown at the top of those pages. Each record's own page lists its revision history under "Last reviewed and what changed".

What this site does not do

• It does not recommend a dose, a schedule, a cycle or a route. Doses on a page are the doses studies administered, reported as published, and the page says so wherever they appear.
• It does not sell, rate, link to or name suppliers, and it carries no affiliate placement.
• It does not translate a study finding into advice for any person. Study populations, species and conditions are stated so a reader can see how far a finding is from their situation.
• It does not present a statement written from general knowledge as a cited finding. Editorial passages are marked as such, and a primary document is added when one exists.

How to read a page

Start with the evidence table: one row per primary study, with design, sample size, species, dose, route, duration and reported outcome. Then read the quoted claims under each heading, checking the tier dot beside each. Open a source from the numbered list to read the passage in context. The reference card at the end summarises the record with its generation date, so a saved copy can be checked against a later version.

Corrections

A record's changelog is the record of what changed and when. Anything found to be wrong is corrected in the record, the correction is dated in the changelog, and the page rebuilds from the corrected record.