How we source

This site has no credentials behind it, so the sourcing rules are the whole of its authority. They are published here in full, including the rubric behind the status badge on every article, so that a reader can check whether we followed them.

1. What counts as a source here

Sources are ranked. Where two of them disagree, the higher one wins and the article says that they disagreed. Where the highest available source is low on this list, the article says that too, in the body, rather than dressing a trade claim up as a finding.

1. Consolidated regulation text and annex entries. The instrument itself, in its consolidated form, read at a stated date — for the European Union that means a CELEX number with the consolidation date printed beside the entry, because annex numbering moves with every amending regulation.
2. Official regulator guidance, notices and FAQ pages — the regulator's own words about its own rule, quoted rather than summarised.
3. Scientific committee opinions — SCCS opinions, CIR safety assessments — cited with the report's own title, year and conclusion wording.
4. Peer-reviewed primary literature with a resolvable identifier, cited with the study type and the number of participants in the sentence that makes the claim, never as a bare “a study found”.
5. Peer-reviewed reviews, used to locate primary work and to describe the state of a field, not to stand in for a paper we have not opened.
6. Trade press and supplier literature, treated strictly as evidence of what the supplier claims and never as evidence that the claim is true. Where an article's only source for a number is trade literature, the article says so on the line.
7. Brand marketing, which is never a source for anything except its own wording — what a pack says is a fact about the pack.

Two rules cut across the ranking. No source may be a summary of a document that has a public address of its own: if the regulation, the opinion or the paper can be opened, we link the thing and not somebody's account of it. And the link text is the resolvable identifier — 10.1021/cr1000145, PMID 32611558, CELEX 02009R1223-20260501, 21 CFR 701.3 — so that a citation survives this site going away.

2. The evidence-status rubric

Every article carries one of three status words in its claim panel, and every one of them links here. The rubric is published because the badge is displayed: a label whose criteria are private is decoration.

Settled

A document is dispositive on its own terms — a legal limit, a printed warning text, a statutory definition, a numbering convention that demonstrably has no standards body — or there is a body of human evidence with no serious dissent. Settled describes the document, not the future: a limit can be amended, and when it is, the change is logged as an update in the corrections log.

Contested

There is real evidence on more than one side, or an observational finding that is genuinely disputed. On a contested page we name who says what, on what data, with what study design and what size, and we report both halves of a finding that has two — including the half the coverage dropped.

Insufficient

Nobody has shown it. The badge carries a date and a sentence saying what evidence would change it. This is borrowed from the Cosmetic Ingredient Review's discipline of concluding insufficient data and treating a use as unsupported until the data arrives. Insufficient is not a synonym for unsafe, and it is not a synonym for safe; it means the question has not been answered.

Why no single number can describe safety

A one-to-ten safety number, a letter grade, a traffic-light rating, a list of ingredients to avoid: each attaches a verdict to a substance, and safety is not a property of a substance in isolation. It depends on concentration, on formulation, on rinse-off versus leave-on, on the site of exposure, and on the individual — five variables that a single number throws away by construction. That is the difference between hazard and risk, and it is the subject of Hazard Is Not Risk: How to Read Any Ingredient Score.

The same objection applies to the vocabulary that usually surrounds a score. Words like “toxic” and “non-toxic” appear on this site as objects of study — claims found on packs and in apps — because in the jurisdictions this site covers they carry no definition a regulator enforces. What they mean, and where the law does bite, is the subject of “Clean” Is a Retail Category, Not a Safety Standard.

3. How we quote the Cosmetic Ingredient Review

CIR reports are one of the few places a reader can find a structured safety conclusion on a named ingredient, and they are easy to quote badly. Our rules:

• The report title in its fixed form — “Safety Assessment of X as Used in Cosmetics” — with the year, so the reader can find the document and see how old it is.
• The conclusion in CIR's own vocabulary, which is a fixed four-way set: safe as used, safe with qualifications, unsafe, and insufficient data. We reproduce whichever wording the report uses instead of translating it into a verdict of our own.
• The qualification text travels with the conclusion. A conclusion of “safe with qualifications” means nothing without the qualifications, which usually name a concentration, a product type or a route of exposure.
• The phrase “as used” is never dropped. It is the load-bearing part of the sentence: the Panel is assessing present practices of use and concentration, not the substance in the abstract.
• CIR's own funding sentence is quoted verbatim, below, so a reader can weigh the source rather than take our word for its standing.

Although funded by the Council, CIR, the Expert Panel for Cosmetic Ingredient Safety, and the review process are independent from the Council and the cosmetics industry.

We print that sentence rather than characterising it. It is CIR's own account of its own independence, the reader can see who pays, and the reader can decide what weight to give a conclusion accordingly. We do not attribute a position to unnamed authorities in the aggregate — a claim is attached to a named committee, a named paper or a named regulator, or it does not appear.

4. Rules for numbers and citations

• A search-volume, traffic or ranking figure is cited to the tool that produced it — including any figure about this site.
• A statistic comes from a paper that was opened. Where only the abstract was available, the article says that on the line.
• A comedogenicity rating names the assay it came from, and the species and site it was run on.
• A consolidated-text annex entry carries its consolidation date. An annex number without a date is not a citation, because the numbering moves.
• Every sentence is written from a document, never from personal use of a product. Nobody here has used the things written about, and the writing keeps to that grammar: what the document says, dated and linked, rather than what a bathroom test found. About sets out why an information site rather than a review site.
• The byline is the publication — a named committee, a named paper or a named regulator carries each claim, and About explains why no person is named.

Where we could not check something

Every article carries a standing section headed What we could not verify, and it is present even when it is two lines. Where a document could not be opened, where a widely repeated figure has no traceable origin, or where a convention is taught everywhere and written down nowhere, that is stated in the body with a date — in those words — rather than smoothed over. Uncertainty is a field on this site, not an absence.

A note on decimals

European Union legal texts use a comma as the decimal separator and put a space before the per-cent sign: an annex entry reads 0,14 %. Inside a quotation, a statute panel or a cell of a table reporting EU text, we reproduce that exactly as written, because it is the document's wording. In our own running prose the same figure is written 0.14%. The two forms are the same number, and seeing both on one page is a feature: the monospace comma tells you that you are reading the law rather than reading us.