Editorial policy

A research claim on this site must trace to a record with a permanent public identifier: a PubMed ID (PMID), a DOI, a ClinicalTrials.gov registration number (NCT), or a Federal Register document number. Regulatory claims trace to the Federal Register or to a document published on fda.gov. We do not cite press releases from vendors, forum posts, influencer content, or secondary write-ups as sources for a research or regulatory claim.

Every cited identifier is machine-checked to confirm it resolves before a page is published, and the whole corpus is re-checked weekly. A source that stops resolving is flagged for review.

A dose may appear on this site only as the dose a specific cited study administered to the population it enrolled, phrased as reporting: “in this 12-week trial of 52 adults, participants received….” We do not publish protocols, cycles, stacks, titration schedules, or any second-person instruction about what a reader should take. Our dose calculators reproduce the arithmetic for a cited regimen and say so on the tool; they contain no “recommended dose” field.

We do not generalize animal doses to people, and we do not describe a compound as treating, curing, or preventing anything unless a cited regulator has approved it for that use.

Research digests and the rolling summaries on compound pages are drafted by an automated pipeline that uses a large language model (currently Claude, by Anthropic). The model is given the abstract or registration record of one study and nothing else, and is instructed to report only what that record says. Its draft is then checked two ways:

• A deterministic check confirms that every claim in the draft is supported by a verbatim quote from the source record, that every dose mentioned appears in the source with the same numbers and units, and that no prohibited language (advice, cures, guarantees, purchasing) is present.
• A second, separate model pass scores how closely the draft follows the source on a 100-point scale and lists any unsupported or prohibited claims. Drafts below 85, or with any unsupported claim, are rejected. A rejected draft may be revised once; if it fails again it is discarded.

The pipeline publishes at most five digests a day and at most two per compound, and it never rewrites a compound profile beyond the clearly marked research summary. Compound profiles and guides are drafted with the same model from a larger set of cited records, checked the same way, and edited by a person. Every page states how it was made in a “How this page was made” note near the bottom.

A person can withdraw any page with a single command at any time. Withdrawn pages return a 404 and drop out of our sitemap within the hour.

Some clinic programs and vendors pay us a commission when a reader signs up or buys through a link marked sponsored. Those relationships are disclosed above the links and explained on How we make money. They have no input into which studies we digest, what a digest says, how compound pages are written, or how clinics are ordered in the directory. No page on this site can be bought.

Vendor pages report what a vendor publishes about itself (products, testing, shipping) together with public enforcement records. Research-chemical vendors sell products labeled “not for human use,” and every vendor page and link carries that label. We check every partner’s site weekly; a vendor that closes or pauses is marked as such on its page the same week, its links are rerouted, and the page stays up so that readers searching for it learn what happened.

Clinic listings are compiled from public map data and refreshed quarterly. A clinic may claim its listing to add a description, the compounds it offers, and a booking link. We verify that the claimant controls the clinic’s email domain or phone number. Claimed listings are marked; they are not ranked higher and cannot pay for placement.

Every research page shows the date it was published and, where different, the date it was last changed. Regulatory status blocks show the date the status was last reviewed; we aim to re-review each compound’s status within 90 days and always when a new Federal Register notice names it.