Method
How the Register works
Every grade on this site means the same thing on every entry, and it means it because it is defined here first. This page is the contract. If a verdict ever seems harsher or softer than the evidence behind it, the fault is in the entry, not in the vocabulary, and the entry is the thing that gets fixed.
Somebody asks a fair question when they land on a verdict here: who decided that, and against what? This page is the answer, written before the entries rather than after them, so it can be held against me.
The bar, written down first
“We found nothing” is worthless unless somebody said in advance what would have counted. So here is what counts:
A human study, published and peer-reviewed, with a comparator — placebo, vehicle, or an active with its own evidence — reporting the endpoint the claim is actually about. In-vitro work, animal models, uncontrolled before-and-after series, consumer-perception surveys and unpublished supplier data are read and reported, but they do not clear the bar on their own.
Work that falls below that bar is not ignored — most entries here spend their longest section on it, because the best available study is usually the interesting one. It is read, described and named. It just does not get to settle the question on its own.
And absence decays. A claim with no trials in September has trials in March, which is why every entry carries the date the search was last run rather than the date the page was last touched. Those are different dates and only one of them is honest.
The six evidence positions
Not letters. A letter grade implies one line with better at one end and worse at the other, and the most important distinction in this whole register — between a claim that was tested and failed and a claim nobody has ever tested — is not a distance along a line. It is a difference in kind. Six positions, and an entry sits in exactly one.
- Established Holds up
-
Controlled evidence, in this population, at this endpoint, pointing the same way more than once. It happens, and when it does it gets said out loud here in the same tone as everything else.
- Partially established Smaller than advertised
-
Supported — for a narrower claim than the one being made. The narrower claim is named in the entry, because "partially" on its own is a shrug and a shrug is not a finding.
- Thinly supported Real, and not enough
-
There is genuine evidence and there is not enough of it, or it is not good enough, to conclude. This is not a verdict against the claim. It is a verdict of not yet.
- Untested as claimed The parts, not the whole
-
The components or the mechanism have support. The specific claim, phrased the way it is phrased on the box, has never been put to a test in that form.
- Not established Studied; does not support it
-
Adequately studied. The evidence does not support the claim as made. This is the strongest thing said here, and it is used sparingly, because most of the time the honest answer is that nobody looked.
- No adequate source found Searched; nothing met the bar
-
Searched, against a threshold published in advance, and nothing meeting it exists. That is a finding about the literature. It is not a statement that the product does nothing, and it should not be read as one.
The nine types of absence
This is the part that took the longest to work out and it is the part worth stealing. “There is no evidence” is the laziest sentence in beauty writing and it is nearly always false. There is usually evidence. It belongs to a different concentration, or a different route into the body, or a different condition, or the company selling the thing paid for it and never published it. Nine names for nine different failures, and an entry gets one whenever it sits below Partially established.
- True absence
-
Nobody has run the study. Not a small study, not a weak study, not a study with the wrong endpoint. None.
- Supplier-generated only
-
The evidence pointing the claimed way was produced or paid for by the people selling the thing. Sometimes that means an in-house study nobody outside the company can read. Sometimes it means published trials whose effect disappears the moment the independently funded ones are read on their own. Industry money does not make a study wrong. It does mean it is not the same evidence as a study nobody paid for.
- Characterisation failure
-
Something was tested, but what was tested is not specified well enough for anyone to reproduce it — source, dose, concentration, particle, batch. The result belongs to one jar, not to a category.
- Claim transfer
-
The evidence is real and belongs to something else: a different route of delivery, a different concentration, a different form of the molecule. It was picked up and carried across, and the carrying is where it broke.
- Category error
-
The cited study is about a different thing that happens to share a name. Common, embarrassing, and it survives because almost nobody opens the citation.
- Plausibility gap
-
The mechanism is sound in a dish. The step where the ingredient reaches the layer of skin that matters, at the amount actually in the bottle, is the step nobody has shown.
- Protocol untested
-
Every ingredient in the routine has support on its own. The sequence, as sold, has never been tested as a sequence — and the sequence is the product.
- Claim inflation
-
The study found something. The claim describes something larger, or faster, or more certain. The distance between the two is usually the marketing budget.
- Indication mismatch
-
There is decent evidence for a condition. It is not the condition the product is sold for, and often not one you have.
What an entry has to carry
The header block is identical everywhere: the claim in the market's own words, the same claim rewritten so somebody could test it, a verdict of twenty-five words or fewer, a position, a type of absence, the regulatory status and what that status does and does not tell you, and the databases, search strings, screening counts and date.
Underneath, four sections that cannot be filled in by pattern. The strongest available study, named, with its design and its funder and whatever is unflattering about it. What would change the verdict — the specific trial, with a population, a comparator, an endpoint and a duration. What the verdict does not mean, because no evidence of benefit and evidence of harm are different findings and letting them blur is the most common way a careful sentence turns into an unfair one. And what you are and are not paying for, which is the question everybody actually arrived with.
Two phrases are banned outright. More research is needed, because it is the fingerprint of every machine-written page this register is competing with and because it occupies the exact slot where a specific study should be named. And scam, along with its relatives, because it describes an intention I cannot see and would have to prove.
There is one more test, and it is the one that actually keeps entries honest: if the strongest-study section, the what-would-change section and the what-you-are-paying-for section could be lifted out and pasted into a different entry without anyone noticing, the entry is not finished. It gets held back rather than published thin.
What this register will not do
It will not name a brand in a verdict. Claims travel across dozens of companies and the claim is the subject here; where a regulator has already ruled on a named company the ruling is reported as the public fact it is, with a link, and that is a different act.
It will not tell you a product does nothing. That is a claim about the world that would have to be proved. What is on offer instead is a claim about my own reading — searched here, on this date, against that bar, found this — which anybody can repeat and contradict.
It will not recommend a purchase and it will not link to a shop. There are no affiliate links anywhere on this site and there is no chair to fill, which is worth checking on anyone telling you what to buy. Including me.
And it is not medical advice. I am not your provider, I have never seen your skin, and nothing here clears you for a treatment or diagnoses anything.
Take the data
Every entry, every grade, every absence type and every search date is available as a file. Cite it, build on it, argue with it — the taxonomy is more useful to everybody if more than one publication uses it, and a bad scrape helps nobody.
CC BY 4.0. Link the entry you took it from.