How this site is made

What every page is built from

Behind every glossary page, comparison, pairing verdict and checker sits one structured file: a knowledge base of 392 ingredient rows, each carrying the same 70 columns, from which 388 ingredient pages are built. Nothing on those pages is written by hand, ingredient by ingredient; a page is a rendering of its row.

What the columns hold, and where each is read from: the name and its INCI name, the family and the job it does (cosmetic-chemistry references); six 0 to 10 scores against that job - efficacy, hydration, barrier support, brightening, acne support, anti-ageing - and an evidence level from strong to weak (the primary literature, graded ashow we score sets out); the irritation and comedogenicity bands, and where one exists the Fulton rabbit-ear rating from the published lists; the pregnancy flag (the regulator's position and the literature, never a forum); whether the ingredient makes skin more vulnerable to UV, with the basis and the source per row, and "unknown" kept as an answer rather than filled in; the time a trial took to show a result, what that trial measured and how the range was read; the effective range the trials worked from, and what the EU limit is and where it is written; photolability; the allergen flags; and the molecular structure, from PubChem.

The base is a file in the repository, reviewed and dated, and the pages are rebuilt from it whole. The last content change to it was on 18 October 2026.

The sources, and how much of each was read

A claim on a page rests on a source recorded in the row it came from. Sources are recorded in 6 columns (the evidence sentence, the effective range, the time to a result, sun sensitivity, photolability, the EU limit), each as a typed reference: a PubMed id, a DOI, an SCCS opinion, a CIR safety assessment, a US federal regulation, an EU regulation or an FDA drug label - and each with a word for how much of it was read: the whole document, the abstract only, or a summary page only. That word is load-bearing. The house rule is that a claim may weaken against its source and never strengthen, and that is unauditable without knowing whether anyone got past the abstract. 288 of the 388 rows carry a source behind their evidence sentence; across the base there are 1,480 source references to 673 distinct documents.

• 701 PubMed records: 49 read in full, 648 the abstract only, 4 a summary page only. Linked to the record.
• 27 papers by DOI: 2 read in full, 20 the abstract only, 5 a summary page only. Linked to the record.
• 69 SCCS opinions (the EU Scientific Committee on Consumer Safety): 8 read in full, 61 a summary page only. Printed as the identifier: the body moves its files, and a dead link is worse than an id you can search.
• 157 CIR safety assessments (the Cosmetic Ingredient Review panel): 4 the abstract only, 153 a summary page only. Printed as the identifier: the body moves its files, and a dead link is worse than an id you can search.
• 6 US federal regulations (21 CFR, the OTC monographs): 3 read in full, 3 a summary page only. Linked to the record.
• 520 EU regulations (Regulation 1223/2009 and its amendments): 18 read in full, 502 a summary page only. Linked to the record.

Where a body keeps a stable address - PubMed, a DOI, EUR-Lex by its ELI, the eCFR by section, DailyMed by setid - the citation is a link to the record itself, and the link is a plain one: a citation to PubMed or the Official Journal is vouching we mean. Where the body does not (SCCS opinions, CIR reviews), the identifier is printed and stops there. The site links to institutions, never to brands or shops; the one exception is a labelled affiliate link at the point of buying, and it carries rel="sponsored".

The regulatory instruments the base reads: Regulation (EC) 1223/2009 and its annexes and amendments (the EU's list of what may be used, at what level); the opinions of the EU Scientific Committee on Consumer Safety that those entries rest on; the Cosmetic Ingredient Review panel's safety assessments; the US over-the-counter monographs in 21 CFR (sunscreens, acne, skin protectants); and, for a prescription ingredient, its FDA label. A safety ceiling is not a clinical finding: the regulator is the right source for it, and the base says so rather than citing a trial that never existed.

The rules, and where the verdicts come from

The one-line verdict on an ingredient page, the answer to "can you use these together", the comparison of two ingredients and the read of a product are all produced by fixed rules that run the same way on every row. The scores and the evidence go in, the sentence comes out; the retinol page and the snail-mucin page are held to identical standards, and nobody's product earns a kinder line. How we score walks through the scoring itself; this page is the whole of what sits under it.

From the same rows: 479 pairing verdicts ("can you mix"), each decided by the same rule set over the two rows' actives - separate nights, a morning and an evening, or nothing to keep apart - and 1,228 ingredient comparisons. The 162 product pages read a label against the base ingredient by ingredient, and the 650 product comparisons are derived from those reads; products get no overall score, for the reason given below.

What is measured, and what is assumed

Some pages print a number that looks certain whatever it actually knows - a calculator prints to two decimal places. So a number here belongs to one of two classes, and the page renders them differently on purpose. Measured: a published figure we can cite, printed as a fact. Assumed: a figure with no clean source that the arithmetic nonetheless needs, printed as an assumption, next to the number it produces, never presented as measured. The sunscreen calculator is the clearest case: the dose every SPF number is established at and the dose people actually apply are both measured per square centimetre, so its headline survives however wrong our estimate of a face's area is - and that estimate is labelled as an estimate on the page. The cost-per-use line on the product pages rests on one dose table of 5 rows (a cream on the face and neck at 2.5 fingertip units for face and neck, a sunscreen at the SPF test dose, a serum, a cleanser and an eye cream at a usual amount), each row labelled measured or assumed with its sources beside the number.

The same honesty runs through the base's "unknown" values. Whether an ingredient makes skin more sensitive to the sun is "yes", "no" or "unknown", and unknown is an answer, not a gap waiting to be filled: a measured "no" and an unmeasured "unknown" are different facts, and the pages never collapse them.

The figures and the 3D

Every diagram on the site is drawn from the site's own data files, never from stock: the score profiles, the dose lines, the layer maps and the hub bars are rendered by scripts from the same base the pages read, so a figure cannot disagree with the page it sits on. The molecular masses and structures come from PubChem, the public database of the US National Institutes of Health; 221 ingredients carry a 3D structure drawn from PubChem's conformer, by its compound id in the row. The layer of skin each ingredient acts on is read from the 500 Dalton rule - a molecule's mass against the threshold below which the stratum corneum lets it through - and from the sourced layer data in the base, with the confidence of each placement carried on the page.

What the site does not do

No overall score for a product: we built one twice, wrote down what would count as success before running either, and both times the result told us to stop (the record). No medical advice and no diagnosis: the pages say what an ingredient tends to do and who tends to react to it, in general; a rash, a pregnancy or a condition is a question for a person who can see your face, and the pages say so where it matters. No sponsored placement anywhere: no brand pays for a sentence, a rank or a mention. Affiliate links exist only at the point of buying - on a product's page, under its read - and every one is labelled and carries rel="sponsored"; there are none inside the glossary, the checkers or the comparisons.

How a page changes

A page changes when its row changes, and the row's history is public: thechangelog lists every change to the base by row and by column, and each ingredient page shows when its own row was last reviewed and what changed since. Dates are real dates from that record, never the day the site was last rebuilt. An essay carries the day it was published and the day a person last re-read it against its sources; the second is not bumped to look fresh.

How to report an error

If a page is wrong, say so: write to Maya, or use "Tell Maya" insideMy Shelf, which sends the note with the page it came from. A correction goes into the base, the changelog records it, and the page rebuilds from the corrected row. The error you report is fixed everywhere the row is read, not on one page.

Who writes this

Every word here is written by one person, by name. Maya Faour is the founder of Glow Simplified and its sole writer. She is not a dermatologist, an esthetician or a cosmetic chemist, and the site says so on every page rather than leaving it to be assumed. Her lane is honest synthesis of the evidence, in plain words, and the discipline of saying what is not known. An architect by training and a certified group fitness instructor, she has spent four years on her own evidence-first routine, which is why this site treats skincare as a habit you can keep rather than a product to chase. She lives in Bucharest.

The scoring engine, the tools and the build are by Sasu Dan Liviu. The knowledge base and every sentence on the pages are Maya's.