How the atlas is checked
Methodology
Everything here is reproducible — including by our critics.
§ 01 · The build gatesA bad claim fails to compile.
Each peptide is a YAML file validated against a Zod schema at build time. The build fails — the change cannot ship — on any of:
- a schema violation (wrong shape, missing required field);
- a citation that references an ID not present in the citation registry;
- a citation key that does not match its own
idfield; - a filename that does not match its slug, or a duplicate slug.
Every claim-bearing value carries a cite slot. When a claim is not yet sourced, the slot is empty and the claim renders as uncited — the absence of a citation is visible, not hidden.
§ 02 · Citation verificationReal ID, and the right paper.
The citation audit (audit-citations.ts) queries live PubMed for every referenced PMID and confirms two things: that the record exists, and that its title matches the title we recorded. The title match is the important part — it catches the most common fabrication pattern, a real PMID attached to the wrong paper. A separate claim-support audit (audit-claims.ts) checks whether a cited abstract actually supports the claim it is attached to; it is being promoted from a reported signal to a hard gate.
None of this is a claim you have to take on faith. The audit scripts are in the public repository. Clone it and run them.
§ 03 · Evidence tiersThe strength of a claim is stated, not implied.
Each claim carries an evidence level on an eight-step scale from fda-approved down through the clinical phases, animal-strong and animal-mechanistic, to theoretical. Each peptide rolls those up into a four-bucket summary tier, computed — never set by hand. Peptides whose evidence base is primarily non-Western literature carry a plain, non-judgemental note saying so, rather than a hidden downgrade.
§ 04 · Lab provenanceFacts about a lab report — never a verdict.
Third-party lab results (HPLC purity, endotoxin) tied to batch codes are being wired in, each independently re-verifiable at the issuing lab. When that ships, the atlas will report facts — “purity as stated: 98.1%,” or “this report code does not resolve at the issuing lab” — and never a verdict like “scam.” Status language runs through a single audited rules layer, and it is not marketed as an independent registry until genuinely arms-length vendors have published into it.
§ 05 · CorrectionsWhen we get it wrong, we log it.
Every material fix — a purged fabricated citation, a corrected statistic, a downgraded evidence rating — is recorded in the public corrections log with its date and its trigger. A reference that turns out to be fabricated is removed and recorded, not quietly overwritten.