Methodology
How this is built, and what it cannot tell you
Every number on this site decomposes into inputs you can check, and every input has a source you can open. Where a figure is a proxy for something we cannot measure, this is where we say so.
Pages
- Evidence grading
- The five inputs, their thresholds, and how a total becomes a letter.
- Extraction
- How fields are read from documents, and the current per-field disagreement rates. Read this one.
- Collection
- Which sources are queried, what rate limits were observed, and what a coverage gap means.
- Trending
- Why nothing is currently trending, and why that is the correct output.
- Vendor transparency
- What the score measures, and the thing it cannot tell you.
- Submissions
- How first-party responses are aggregated, and how adverse reports are routed.
- Publication status
- What is graded, what is published, and what is still gated.
- Corrections and removal
- How a voice, vendor, or rights holder gets something changed or taken down.
- Vendor right of reply
- Contest an input behind a transparency score, with the document that shows it is wrong. Recorded and reviewed, never applied automatically.
- Correction or removal of a mention
- If a mention attributed to you is wrong, mischaracterises your position, or should come down.
Source tags
Every record carries exactly oneTags are ordered, and the order is enforced rather than encouraged: nothing below REGULATORY appears above HUMAN-TRIAL on any surface of this site, including search results. The four visual treatments below map to how much a record can be relied on — solid ink for primary evidence, a seal for agency filings, an outline for commentary with no data behind it, and hatched ochre for anything published by someone selling the product.
- HUMAN-TRIAL
- Peer-reviewed human study.
- ANIMAL
- Peer-reviewed in vivo animal study.
- PRECLINICAL
- In vitro, mechanistic or computational work — or, for most records carrying this tag, work whose study design could not be determined. The tag is the floor of the evidence hierarchy, so an unclassified paper cannot outrank a real trial.
- REGULATORY
- Agency filing, warning letter, alert or adverse event report.
- VENDOR-CLAIM
- Published by an entity selling the product.
- SELF-REVIEWED
- Review hosted by the seller being reviewed.
- INDEPENDENT-REVIEW
- Third-party platform or community review.
- EXPERT-COMMENTARY
- Podcast, newsletter or clinician commentary. No primary data.
- MEDIA
- News and trade coverage.
- USER-REPORT
- First-party submission through our form.
Standing limitations
- Judgment fields are not human-audited. Study type, sample size, blinding, preregistration, funding, direction of effect, outcome mapping and mention stance are extracted by automated agents. Current disagreement rates.
- Nothing here verifies what is in a vial. We do not test products. Reviews detect a narrow band of failure and do not detect underdosing, substitution, or degradation, because “no effect” is indistinguishable from a compound that does not work.
- A grade is not a safety assessment. It measures how much evidence exists for one outcome. A grade A pair can have a serious side-effect profile and a grade E pair can be harmless; neither is what the letter is about.
- Coverage is incomplete and the gaps are recorded. Where a source blocked us, rate-limited us, or could not supply the window we asked for, the gap is on the peptide record rather than worked around.
- Mentions are attention, not evidence. They never feed a grade or a transparency score.