Why an ingredient is in a formula at all
If an ingredient is in a blend, it is because the evidence found for it was judged strong enough to justify carrying it. That is an editorial judgment, and it is stated here as one rather than dressed as a finding.
The search behind that judgment is not editorial. It runs in every language the field publishes in, and continues until new sources stop changing grades — because “we found no human trial” and “no human trial exists” are different statements and only one of them can be earned. 108 of the 1,792 pages published here carry a source in a non-Latin script — Chinese, Japanese, Persian and Russian — quoted verbatim in the original alongside its identifier. The method, with a worked six-language example, is published at How We Search.
What an exhaustive search cannot do is manufacture evidence that does not exist. Where the best available evidence for an ingredient is a cell or animal study, that is what the grade says — which is why 506 of 946 ingredients carry grade B. The grade reports what the search found, not what we would have preferred it to find. Everything below is that reporting.
What this catalog does not have
Most catalogs tell you what they have. This page tells you what they do not, with a number against each line, because a claim is only as good as a reader’s ability to find its edge. Every figure here is recomputed from the corpus by a build guard, and the build fails when a checked figure drifts. 47 guards and 3 cross-surface gates run on every build; what they do and do not cover is stated below.
| What a reader may want | Counted |
|---|---|
| Ingredients whose pathway rests on randomized human trials (grade A) | 104 of 946 |
| Ingredients corroborated by human data or independently replicated animal work (grade B+) | 291 of 946 |
| Ingredients resting on cell or animal evidence only (grade B) | 506 of 946 |
| Ingredients resting on traditional or historical use only (grade C) | 45 of 946 |
| Ingredients carrying no grade at all | 0 of 946 |
| Finished blends tested as a complete formula in a clinical trial | 0 of 68 |
| Independent third-party audits of the grades published here | 0 |
| Blends with a published outside-laboratory certificate | 1 of 68 |
| Pathway pages, each carrying its own citations | 567 |
| Cited studies resolving to a DOI, PMID or PMC identifier | 3,497 |
Read the first four rows together. Fifty-three percent of the ingredients in this catalog rest on cell or animal evidence and have not been shown to do the same thing in people, and a further 45 rest on traditional or historical use alone. That is printed in the grade beside every claim on every one of those 506 pages. It is the single most important number on this site.
What is measured, what is inferred, and what has not been tested
Three different things get called “evidence” on supplement sites, usually without distinction: a trial that measured the endpoint, a trial that measured something next to it, and no trial at all. Each one gets its own tier here, and the middle one gets a number.
A randomized human trial, or a meta-analysis of them, measured this pathway’s own endpoint. That is grade A, and it is the only tier where the published study tested the thing the grade is about. 104 of 946 ingredients hold it.
A human trial exists and is cited, but it measured a different endpoint than the pathway the grade is for. The trial is real; the link from that result to this pathway is a reasoned inference, not a measurement. That is grade B+, and it is why B+ opens a pathway only alongside an independent corroborating driver rather than on its own. Measured on this build: 189 graded rows cite a randomized human trial, a placebo-controlled trial or a meta-analysis while carrying a grade below A — 162 at B+ and 27 at B — for exactly this reason.
For those 189, the confirming trial — one measuring this pathway’s own endpoint — does not yet exist. Not that it failed, and not that it was ignored: the worldwide, all-language search described at How We Search did not find one, and a refusal is recorded as a grade rather than as silence. If one is published, the grade moves and the change is recorded.
A grade here answers a narrower question than the word usually carries: not whether an ingredient has evidence behind it, but whether a trial measured the endpoint the grade is about. The 189 appear here as a figure because a reader cannot audit what a site declines to count.
What the published ratings do not include
Every rating on this site is five stars, and that is not because every customer was delighted. No review below five stars appears because the brand has none: dissatisfaction is settled by refund rather than recorded as a rating. Nobody’s money is kept unless they are happy with the formula, so a customer who was not happy was refunded and left no rating behind. That is a real filter on the 1,178 published reviews, and it is counted here rather than left for a reader to infer from an unbroken run of fives. A rating average drawn from a population that excludes the dissatisfied is not a measure of how often the formula works.
What grade A does not mean
Grade A means a randomized human trial, or a meta-analysis of them, measured this pathway’s own endpoint. That endpoint is usually a biomarker. A trial showing an ingredient lowers hs-CRP is grade A evidence about hs-CRP. It is not evidence that a person taking it will feel better, live longer, or recover from anything. The distance between a moved biomarker and a changed life is real, it is frequently large, and nothing on this site closes it.
What is not disclosed
Ingredient quantities inside a blend are not individually listed. A reader who wants to compare a study’s dose against what is in the jar cannot do so from this site. Every botanical in every blend is named, the concentration ratio is stated and the total weight is stated, but the milligram breakdown is withheld, and that is a real limit on what the evidence here can tell a buyer.
Who assigned these grades
One person: Gavin Robert McGowen, founder and formulator, applying the published rubric at How Evidence Is Graded. There is no advisory board, no external reviewer and no institutional affiliation behind the grading. The rubric is public so a disagreement can be specific; the grades are self-assigned and should be weighed as such.
What safety information exists, and what has never been swept
The measured position is set out here in full.
Measured on this build: of 68 blends, 4 carry a product-specific safety warning a reader can act on, and 64 carry none. An earlier build counted 21 here; that figure came from a keyword match that was also counting the word “pregnancy” inside customer reviews, and it overstated this by more than fivefold. The guard now counts warnings, not words. Of 946 ingredient pages, 24 carry an interaction or contraindication signal, and seven of those are the named-hazard disclosures set out below, written onto the ingredient receipt for each compound so the caution sits beside the dose. Every figure in this section is recomputed from the shipped files by a build guard, so it cannot drift from what the pages actually say.
A systematic drug-interaction sweep across all 941 ingredients has never been run. Nothing here should be read as one. The absence of a note on an ingredient is not evidence that the ingredient is inert — it is evidence that nobody has looked, and this site will say so rather than let silence imply safety.
Podophyllotoxin and its source plant Sinopodophyllum hexandrum, carried in Force Field, are cytotoxic antimitotics — the chemotherapy agent etoposide is derived from podophyllotoxin — and the compound is documented to cause nerve, marrow, liver and kidney injury when swallowed in quantity, and is contraindicated in pregnancy. Yohimbine, carried in Lipolytic, raises blood pressure and heart rate, and is contraindicated with hypertension, a heart-rhythm disorder or an anxiety disorder, and alongside MAO inhibitors. Anyone taking either formula, and anyone pregnant, on cardiac or psychiatric medication, or facing surgery, should put the full ingredient list in front of a prescriber before starting.
Berberine — Inhibits CYP3A4/P-gp (can raise levels of co-taken drugs) and may lower blood sugar (additive with diabetes medication). Avoid in pregnancy, nursing and infants. Licorice — With chronic or high-dose use: raised blood pressure, low potassium and fluid retention (pseudoaldosteronism). Caution with hypertension, heart or kidney disease, or pregnancy. Ashwagandha — May raise thyroid-hormone levels and stimulate immune activity; rare reports of liver injury. Avoid in pregnancy; caution with thyroid or autoimmune conditions and sedatives. Andrographis — May stimulate immune activity (caution with autoimmune disease or immunosuppressants) and has antiplatelet effects. Avoid in pregnancy. Each ingredient receipt states how many blend pages carry that ingredient, computed from the shipped files.
Botanical concentrates at 200:1 are biologically active and can interact with prescription medication, particularly anticoagulants, blood-pressure, blood-sugar, thyroid and immunosuppressant drugs, because ingredients in this catalog act on those same pathways. A pharmacist or prescriber holding a person’s actual medication list can be specific in a way this site cannot, and that is the correct place for the question until the sweep is run and published here.
What happens when new evidence arrives
The search does not stop when a page is published. New literature is sought continuously, both the kind that confirms a grade and the kind that contradicts it — and the second kind is the reason The Evidence Against Us exists. When a study contradicts a grade carried here, it is published there in full, quoted verbatim, with its identifier, so a reader can check it without our summary in the way. Six conflicts are logged there now: three resolved, three still open and labelled as unresolved.
No grade on this site has ever been moved on a single finding, and that is deliberate. One paper is not a weight of evidence. A catalog that pulled an ingredient every time one study disagreed would be responding to noise, and would also quietly delete the disagreement instead of publishing it. A grade moves when the body of evidence moves — and one of the six cases above is marked directly opposes a live grade and left standing, with the reason given, rather than hidden.
What happens on a single finding is this: it gets published, labelled, and the site states plainly whether it changed a grade or did not, and why. A register with no open items would mean we had stopped looking.
Where the studies disagree
They often do, and when they do we publish both sides rather than the convenient one. The clearest worked example is in The Evidence Against Us, where a 2025 meta-analysis of berberine is quoted on both counts: it found no significant effect on HDL cholesterol, systolic or diastolic blood pressure — which narrows a claim this site makes — and in the same paper found berberine does significantly reduce triglycerides, fasting plasma glucose and waist circumference, which supports one. Both halves are on the page, under the heading that exists to argue against us.
That is the standard: where one study says a thing and another says the opposite, a reader gets both and decides with their clinician. We are not the right party to resolve a live scientific disagreement on a sales page, and we do not try to.
What the verification file does and does not prove
/verify.txt publishes a SHA-256 for every machine-readable file served here. It proves the corpus has not changed since publication. It is generated by this site, about this site, and it does not speak to whether the content is correct. It is an integrity check, not an accuracy check, and should never be read as the latter.
Where to find evidence that is not ours
For any ingredient here, the independent sources are better starting points than this site: NIH Office of Dietary Supplements fact sheets, Cochrane reviews, and the primary trials themselves, every one of which is linked by identifier on the page that cites it. This catalog is a map of that literature with grades attached. It is not a substitute for it, and a reader who checks the citations and reaches a different conclusion is doing the thing the citations are there for.
The same register is published machine-readably inside llms.txt, and the two are held to the same figures by a build guard.