Where the data comes from

Two sources, kept distinguishable rather than merged into an anonymous blend. Every field records which one it came from, because “it was in the spreadsheet” is a real answer to “why do you say that” and pretending otherwise helps nobody.

• A curated internal library — summaries, mechanisms, dosing models, benefits and reconstitution presets, plus the gap ledger.
• An operator-verified spreadsheet — regulatory status, routes, and 1010 citations whose identifiers were checked against the source registries.

What is deliberately excluded, and why

This section matters more than the one above it.

• A table of 97 “evidence claims” from the source database. 68% of it was scraped page furniture — timestamped page headers, navigation strings, and in several rows our own product documentation — tagged as human observational evidence. It is not filtered when the page renders; it never enters the export at all, because data that cannot be published should not sit where someone might later switch it on.
• 158 rows that were never compounds. Seller page titles ingested as records: HTTP error pages, order-tracking screens, blog posts, product listings. They were published as indexed compounds and carried 1,227 declared gaps between them — 81% of a ledger we advertised. They are gone, and the count on this site dropped accordingly.

Both exclusions cost us headline numbers. An earlier version of this site said “204 compounds” and “1,515 declared gaps”. Neither survived being checked, so neither survived.

How a rung is assigned

Automatically, from the study design named in each citation, and the derivation is recorded. The grader is built to be precise rather than complete: a citation whose design it cannot read is left ungraded rather than guessed at, because a wrong rung is a false statement on every page that renders it while a missing one is merely missing. 524 of 1010 citations are currently ungraded for that reason.

A compound takes the strongest rung any of its citations earned — never an average. Three animal studies do not add up to a human trial. The weaker citations stay listed underneath in rung order, so the whole basis is visible. The full rubric defines what qualifies for each rung and what each one does not license anyone to conclude.

What “not on file” means

That we looked for a specific value on a specific record and did not find one meeting the bar. It is a claim, not a blank. A blank field invites you to supply the answer from context; a named absence tells you not to. 521 of them are browsable by field and by record.

It does not mean the value does not exist anywhere. It means we do not hold it, and the reason is stated — usually either that no source on file holds it, or that the reference database returned nothing.

How to challenge a grade

Every grade names the phrase it was derived from, so a disagreement is usually about a specific sentence rather than about judgement in general. That is deliberate: it makes the conversation short.

1. Find the claim on the record and note the rung and the phrase it cites.
2. Check it against the rubric — the qualifying and disqualifying criteria for each rung are published in full.
3. Tell us what the rung should be and which source supports it. A citation beats an argument; we will change a grade for a paper and not for an opinion.

Corrections are made to the record and the rubric's revision history notes any change to the method itself.

What this site is not

It is not a clinician, it does not recommend a compound or a dose, and it does not sell anything. See the disclaimer. It is also not complete, and the gap ledger is the honest measure of how far from complete.