Methodology

The numbers, with definitions

The big numbers on the homepage, the about page and this page used to be typed by hand. So “human studies” and “citations” could both be true and still disagree. Now one piece of code counts them all. Each label below is the definition it uses.

What gets included

We report the evidence that exists at every level, and we label the level. We do not treat government approval as the final word on what is true. A page here should say what was measured in people first, with numbers. Then it should say exactly what has not been shown. Where the record is thin, it should say why.

1. 01

   Not approved is a legal fact, not a finding about whether it works.

   A compound can lack approval because a study failed, or because nobody ever ran one. Those are different things, and every page says which.

2. 02

   No big drug study does not mean no results in people.

   Small safety studies, studies that follow large groups of people, clinic reports, studies of a closely related molecule, and what users report are all results from real people. We report them, labelled for what they are, instead of writing "none".

3. 03

   Where nothing human exists, we say so precisely.

   "No study in people of any kind" means exactly that, and we name where we looked.

4. 04

   Why a trial is missing is part of the answer.

   Nobody pays for a 300-person study of a molecule nobody can patent. So when there is no study, that tells you how drug research is paid for, not whether the molecule works. It cuts both ways: with no study, a real harm and a real lack of benefit would both go unseen.

5. 05

   Negative results stay on the page.

   If a study found nothing, we say so and give the numbers. We remove wrong dismissals, not results we do not like.

6. 06

   Open disputes are labeled as disputes.

   Where research groups disagree about a result, we say the question is contested rather than picking a side and presenting it as settled.

How sources are checked

1,587 checked sources across 453 compound pages. Those numbers are counted from the database, not typed in by hand.

1. 01

   Results in people come first, with numbers.

   A page leads with what happened to people: who they were, how many, what they were compared with, and how big the change was. Animal and human results never share a paragraph, and where the result is in mice the sentence says mice.

2. 02

   Every study is looked up live before we write it down.

   Each PubMed ID goes to the NCBI E-utilities record, each registration number to the ClinicalTrials.gov API, and Federal Register documents to the live HTML. The returned title is compared with the title on our page, the year is compared with the publication date, and the record is scanned for retraction markers. Nothing is written from memory.

3. 03

   A separate pass re-checks the citations after the fact.

   An independent script walks every citation in the database, resolves it again at the source, and exits with an error on any mismatch. Fabricated citations have been caught on this project six times, including ones we produced ourselves. The re-check exists because of that, and it is the reason this page can make the claim it makes.

4. 04

   A validator runs over the whole database before anything ships.

   It fails on a duplicate compound, a link to a profile that does not exist, a missing legal status, a citation year outside 1950 to 2026, a citation URL that is a search query rather than a record, and a review date that does not parse. It also fails on an em dash, which is a house style rule enforced in code so it cannot drift back in, and on any sentence that claims the site takes no commission while price rows exist.

5. 05

   Every page carries the date it was last reviewed.

   The record moves. A profile that has not been looked at since last spring says so, and when a new finding changes what a page should say, the old wording is corrected in public with a date rather than quietly replaced.

The floor for famous names

Gaps are findings too

Archive rating

Sellers on the price pages carry a certificate tier and, beside it, an Archive rating. The rating is two paths. Source-backed is what the storefront itself shows. Archive evidence is what this archive checked. Each path is capped at 4, and the caps are printed. The paths are not added into one hidden total. A higher price does not raise either path. The same text is on the certificate search.

How this is funded

Knowledge stays free. Every page, every guide and every tool is free to read and stays free.

The archive may use disclosed commercial links. It does not sell peptides, take sponsorships, or sell placement.

We never charge for knowledge. No vendor has a say in what a page says. The harder test is whether a page tells you what went against it. No sponsored compounds, no paid spots, no ads, no discount code of our own, no supplement line. Sellers on price pages are ranked by their lab certificate first and their price second. Any page is free to tell you not to buy at all.

The live price board is /prices. Every test paper we could open is at /certificates. The Archive rating next to the certificate grade runs on two separate paths, each capped at 4, shown in the section above. Paying more never lifts either one. The labs named on seller certificates are at /prices/labs, with how a buyer can look a certificate up. That page does not make up purity test scores or a letter grade for labs. The purity and dose filter is written but switched off. It stays off until we have quality data we have the right to use.

Refresh and review dates

Open data

The same counts, plus one row per compound, are free to download or pull by code at /data. Licensed CC BY 4.0, so you can reuse it if you credit us. No personal data and no sales tracking links. For any medical number, cite the compound page.

What we refuse to count

1. 01

   Sequence databases.

   UniProt, NCBI protein records and similar catalogues count sequences, not human evidence. A sequence is not a study, and we do not add it to a citation total.

2. 02

   Vendor SEO farms.

   Dosing blogs attached to storefronts, affiliate roundups, and purity pages that cite each other. They are not primary literature. When a vendor page is mentioned on a profile, it is labelled as a source of a dosing convention or a side-effect category, not as evidence that anything works.

3. 03

   Invented match scores.

   No compound is scored as best for an outcome. Participant counts, trial counts and null counts are arithmetic over this catalogue. They understate the wider literature and they say so.

4. 04

   First-hand reports added into a grade.

   What people using a compound report is counted separately from human studies and never added into the study total. A thousand people saying a thing worked is not a trial.