Corrections

Every comedogenic number on this site is a claim about a printed row in a specific paper. When one of those claims turns out to be false, we kill it, and we write down what it was and what killed it. This page is that record.

It is generated from data/retractions.json, the same file the build checks every page against. If one of these claims reappears anywhere on this site, in any wording, the site does not compile. A retraction that does not propagate is not a retraction.

Numbers in circulation that no document supports

These are the claims you will meet everywhere else. Each is followed around the internet by a number, and in each case we went and read the paper the number is supposed to have come from.

  1. We said

    phenoxyethanol's circulating comedogenic 0 is Fulton's phenoxyethyl paraben row read across, and that the chart-makers read that row.

    Fulton 1989 does print 'Phenoxyethyl paraben — 0' and does not print phenoxyethanol, so the near-identical name carrying the same number is a live candidate. A candidate is not a chain. We never traced one, and we asserted the route anyway — in prose on the propylparaben page and, worse, in legacy-assays.json, which is the file the rest of the site reads its provenance from. This is the same move as the sulfated castor oil jojoba story, retracted on 2026-08-12, where the tidy substitution turned out to be a guess and the material had a printed row of its own all along.

    2026-08-26 · Found by: The family-hub fact-check batch, 2026-08-26. The preservative hub refused the claim; two member pages and our own reference data were still asserting it.

  2. We said

    Fulton's 'Stearic acid: TEA - 3' is a clean reading of the isolated TEA-stearate soap - an amine, an acid and the salt they make together, each measured on its own - so that the 3 is the salt's own comedogenic score, a grade above the plain acid's; and that the CIR's 1995 component read-across is the methodological opposite of Fulton measuring the isolated salt.

    Fulton's own text reports the entry as a formulation experiment: ratios of stearic acid to triethanolamine (4:1, 1:1, 1:4) tested in a cold cream base, all combinations comedogenic. That is evidence about mixtures in an incompletely described base, not an assay of a defined isolated salt; the 3 is the top of plain stearic acid's printed 2-3* range, not a grade above it; and the CIR's 1995 final report records that published data on TEA Stearate as an individual ingredient were not available, so no isolated-material reading of the salt exists in the corpus we hold. A safety read-across and a comedogenicity formulation test are different endpoints and different evidentiary objects, not opposite answers to one question.

    2026-08-26 · Found by: External fact-check of the emulsifier family hub, 2026-08-26. The isolated-salt framing was live on the published triethanolamine and stearic acid pages, corrected in the same commit as this entry.

  3. We said

    Fulton's two cetyl alcohol readings, 2 in 1989 and 3 in 1984, are both borderline by his 1989 key.

    Each paper published its own key and they are not the same key. Fulton 1989: 0-1 not significant, 2-3 borderline, 4-5 considered positive. Fulton, Pay & Fulton 1984: 1 to 2 not significant, 3 or above considered positive. The 1984 three is POSITIVE in the paper that printed it. Reading it under the later paper's key demoted a positive result to a borderline one, and the page carried a Clear rating on the strength of it. This was in the meta_description, so it shipped. Rating moved to Caution on 2026-08-20.

    2026-08-20 · Found by: The cross-paper conflict sweep, 2026-08-20. The same sweep found five more published pages rated Clear on a reading inside the borderline band.

  4. The claim

    tea tree oil has a comedogenic rating of 1.

    No document in the corpus assigns tea tree oil any comedogenicity score. Fulton 1989 (all 220 rows), Fulton 1984, Morris & Kwan 1983 and Nguyen 2007 contain no tea tree, Melaleuca or Melaleuca alternifolia row. Unlike most of the numbers on this site, this one is not even a misread of a neighbouring row — there is no row to misread. It names no study, no scale and no year.

    2026-08-12 · Found by: Adversarial fact-check, 2026-08-12, checked against all four primary inventories.

  5. We said

    octyl palmitate has a measured comedogenic dose-response — scoring 2 neat, 1 at 50% and 0 at 5% — reported by a named study.

    We printed that three-point series and attributed it to Draelos & DiNardo (2006). It is in neither paper it could have come from. Draelos & DiNardo tested FINISHED PRODUCTS on human backs and print no per-ingredient dilution series at all. Mills & Kligman 1982, the other candidate, works on a 0-3 scale, and the series as we printed it was explicitly 'on the 0-to-5 scale' — so it cannot be theirs either. We deleted the series outright rather than re-attributing it, because we have not read it in any primary document (D56). What we CAN carry for octyl palmitate is Fulton 1989's row of 4 on the 0-5 scale, and Waranuch 2021, in which octyl palmitate was the POSITIVE control on the upper backs of 15 healthy men and raised comedone counts from 6.1 to 27.3 over four weeks. Neither is a dilution curve. There is no dilution curve.

    2026-08-05 · Found by: The 2026-08-05 adversarial fact-check flagged the attribution; verification against legacy-assays.json killed the fallback attribution too, when the Mills & Kligman scale (0-3) turned out to be incompatible with the scale our own sentence named (0-5).

  6. We said

    Draelos & DiNardo (2006) report a comedogenic score for an individual ingredient.

    They do not, and the shape of their paper is the reason it matters. 'A re-evaluation of the comedogenicity concept' tested FINISHED PRODUCTS on small human panels and found products testing non-comedogenic despite containing ingredients the rabbit assay had scored. It is a critique of the method. Citing it for any per-ingredient number inverts the paper: the one study we hold that says ingredient scores do not predict product behaviour cannot also be the source of an ingredient score. We are allowed to cite it for the caveat, which is what it actually establishes.

    2026-08-05 · Found by: Chasing the octyl-palmitate dilution series back to the paper it was attributed to, and finding the paper contains no such table.

  7. The claim

    Fulton 1989 scored rosehip oil 0 to 1.

    Fulton's table has NO ROSEHIP ROW. Not Rosa canina, not rosa mosqueta, not rose hip. We rendered Table I at 400dpi and transcribed all 220 rows by eye: the string 'rose' occurs in it exactly ONCE, inside 'Evening primrose oil' (graded 3) — which is Oenothera biennis, a plant in the Onagraceae, not a rose at all. This is not even the usual failure of misreading an adjacent row. There is no row to misread. The number was attached to the paper by somebody who never opened it.

    2026-07-13 · Found by: Our own draft, from reading the primary table. The number circulates as 0-1, 1, 1-2, 2 and even 3-4 depending on the chart — which is itself the tell: a real measurement does not come in five values.

  8. The claim

    sodium lauryl sulfate scores 5 for comedogenicity.

    PARTLY supported, and this wording is corrected as of 2026-07-18 — it used to open 'Supported by NO document,' which is now false. Fulton's 1984 rabbit table (J Am Acad Dermatol 10:96-105, read in full off the journal scan) prints Sodium lauryl sulfate = 5, used 'as is'. So the flat 5 the charts print is Fulton's own 1984 reading, NOT invented. It is still one reading in a collision: Fulton, Bradley, Agueda & Black 1976 (Cutis 17:344), via CIR 1983, gives 3 at 1% and 4 at 5% (moves with concentration); Morris & Kwan 1983 scored it 0 at both 5% and 10% and wrote that they conflict with Fulton; Fulton's 1989 table has no SLS row at all. Four rabbit results that do not agree — 5, 3/4, 0, no-row — and the internet flattened them into one settled 5. A bare flat 5 presented as THE answer is still wrong because it hides M&K's 0 and the concentration-dependence; but 'no document supports it' is no longer true. Guard kept to stop a settled-5 claim; the SLS page (allowlisted) tells the four-way story.

    Also a CORPUS finding: Fulton 1976 is a NINTH numeric paper our provenance map had never heard of. Every time we have said 'there are only N documents with comedogenic numbers in them', N has been wrong. legacy-assays.json now says to treat the corpus as OPEN.

    2026-07-12 · Found by: External fact-check, sodium lauryl sulfate — it ran down a source we had given up on.

  9. The claim

    acetylated lanolin scores 4 / is strongly comedogenic.

    Fulton 1989 Table I prints 'Acetylated lanolin 0 0' — zero comedogenicity, zero irritancy. The 4 belongs to ACETYLATED LANOLIN ALCOHOL (a different molecule, different CAS: 61788-49-6 vs 61788-48-5), and Fulton's own prose says even THAT is 'not because of the acetylated lanolin but because of the cetyl acetate additive' — cetyl acetate has its own row at 4. Lubrizol still sells the material as 'cetyl acetate and acetylated lanolin alcohol'. Nearly every checker on the internet has dropped the word ALCOHOL.

    UPDATE 2026-07-18: the 1984 print copy is now obtained and READ in full. Fulton, Pay & Fulton 1984 Table I prints 'Acetylated lanolin — 4' — for the plain ester, with NO separate 'acetylated lanolin alcohol' row (that split is a 1989 refinement). So Fulton's own two papers CONFLICT on the plain ester: 4 in 1984, 0 in 1989. The claim 'acetylated lanolin scores 4' is therefore Fulton-1984-SUPPORTED and Fulton-1989-CONTRADICTED — a same-author conflict, not a clean debunk. The 1989 correction (on THAT table the 4 is the alcohol's) still stands, so the guard stays to stop a bare 'acetylated lanolin is a strong 4' claim; but the honest story is now a conflict, and the acetylated-lanolin page (allowlisted, rated Caution) carries both readings. Prior wording, kept for the record: Fulton's 1984 abstract says 'Lanolins continue to be a problem, especially derivatives such as ACETYLATED or ethoxylated lanolins.'

    2026-07-12 · Found by: Our own re-review, D69, after Jake flagged that every other tester lists it as clogging.

  10. The claim

    Fulton 1989 assigned a comedogenic score to coconut oil.

    Fulton's Table I has NO coconut oil row. He tested 'Coconut butter' (4). The most-quoted comedogenic score in skincare is attached to the wrong material.

    2026-07-12 · Found by: External fact-check, coconut oil, round 1.

  11. The claim

    sodium chloride scores 5 for comedogenicity.

    No primary document assays sodium chloride. It has no row in Fulton 1989, none in Morris & Kwan 1983, and none in any numeric paper we hold. The 5 is not a measurement. WE ALSO GOT ITS ORIGIN WRONG, and that correction belongs here rather than quietly in a commit: we said the 5 traced to a 2014 user comment on CosDNA. It cannot. The number was already being quoted on the acne.org forum on 14 September 2005 — nine years earlier — in a thread that opens "How come sodium chloride (salt) has a comedogenic rating of 5 out of 5 (worst)" and in which nobody could source it either. We asserted a citation chain we had not followed, which is the exact failure this ledger exists to catch, committed inside the ledger. Where the 5 came from is still unknown, and we are not going to guess a second time.

    2026-07-11 · Found by: The material-identity check found there was no assay. Our own sodium-chloride draft then found that our stated ORIGIN for the number was false, by going and reading the 2005 forum thread we had never opened.

  12. The claim

    wheat germ OIL scores 5.

    The paper prints wheat germ GLYCERIDE at 3. No oil, in any study, ever scored 5.

    2026-07-11 · Found by: The material-identity check.

When we get it wrong

  1. We said

    phenoxyethanol's circulating comedogenic 0 is Fulton's phenoxyethyl paraben row read across, and that the chart-makers read that row.

    Fulton 1989 does print 'Phenoxyethyl paraben — 0' and does not print phenoxyethanol, so the near-identical name carrying the same number is a live candidate. A candidate is not a chain. We never traced one, and we asserted the route anyway — in prose on the propylparaben page and, worse, in legacy-assays.json, which is the file the rest of the site reads its provenance from. This is the same move as the sulfated castor oil jojoba story, retracted on 2026-08-12, where the tidy substitution turned out to be a guess and the material had a printed row of its own all along.

    2026-08-26 · Found by: The family-hub fact-check batch, 2026-08-26. The preservative hub refused the claim; two member pages and our own reference data were still asserting it.

  2. We said

    Fulton's 'Stearic acid: TEA - 3' is a clean reading of the isolated TEA-stearate soap - an amine, an acid and the salt they make together, each measured on its own - so that the 3 is the salt's own comedogenic score, a grade above the plain acid's; and that the CIR's 1995 component read-across is the methodological opposite of Fulton measuring the isolated salt.

    Fulton's own text reports the entry as a formulation experiment: ratios of stearic acid to triethanolamine (4:1, 1:1, 1:4) tested in a cold cream base, all combinations comedogenic. That is evidence about mixtures in an incompletely described base, not an assay of a defined isolated salt; the 3 is the top of plain stearic acid's printed 2-3* range, not a grade above it; and the CIR's 1995 final report records that published data on TEA Stearate as an individual ingredient were not available, so no isolated-material reading of the salt exists in the corpus we hold. A safety read-across and a comedogenicity formulation test are different endpoints and different evidentiary objects, not opposite answers to one question.

    2026-08-26 · Found by: External fact-check of the emulsifier family hub, 2026-08-26. The isolated-salt framing was live on the published triethanolamine and stearic acid pages, corrected in the same commit as this entry.

  3. We said

    Fulton's two cetyl alcohol readings, 2 in 1989 and 3 in 1984, are both borderline by his 1989 key.

    Each paper published its own key and they are not the same key. Fulton 1989: 0-1 not significant, 2-3 borderline, 4-5 considered positive. Fulton, Pay & Fulton 1984: 1 to 2 not significant, 3 or above considered positive. The 1984 three is POSITIVE in the paper that printed it. Reading it under the later paper's key demoted a positive result to a borderline one, and the page carried a Clear rating on the strength of it. This was in the meta_description, so it shipped. Rating moved to Caution on 2026-08-20.

    2026-08-20 · Found by: The cross-paper conflict sweep, 2026-08-20. The same sweep found five more published pages rated Clear on a reading inside the borderline band.

  4. We said

    octyl palmitate has a measured comedogenic dose-response — scoring 2 neat, 1 at 50% and 0 at 5% — reported by a named study.

    We printed that three-point series and attributed it to Draelos & DiNardo (2006). It is in neither paper it could have come from. Draelos & DiNardo tested FINISHED PRODUCTS on human backs and print no per-ingredient dilution series at all. Mills & Kligman 1982, the other candidate, works on a 0-3 scale, and the series as we printed it was explicitly 'on the 0-to-5 scale' — so it cannot be theirs either. We deleted the series outright rather than re-attributing it, because we have not read it in any primary document (D56). What we CAN carry for octyl palmitate is Fulton 1989's row of 4 on the 0-5 scale, and Waranuch 2021, in which octyl palmitate was the POSITIVE control on the upper backs of 15 healthy men and raised comedone counts from 6.1 to 27.3 over four weeks. Neither is a dilution curve. There is no dilution curve.

    2026-08-05 · Found by: The 2026-08-05 adversarial fact-check flagged the attribution; verification against legacy-assays.json killed the fallback attribution too, when the Mills & Kligman scale (0-3) turned out to be incompatible with the scale our own sentence named (0-5).

  5. We said

    Draelos & DiNardo (2006) report a comedogenic score for an individual ingredient.

    They do not, and the shape of their paper is the reason it matters. 'A re-evaluation of the comedogenicity concept' tested FINISHED PRODUCTS on small human panels and found products testing non-comedogenic despite containing ingredients the rabbit assay had scored. It is a critique of the method. Citing it for any per-ingredient number inverts the paper: the one study we hold that says ingredient scores do not predict product behaviour cannot also be the source of an ingredient score. We are allowed to cite it for the caveat, which is what it actually establishes.

    2026-08-05 · Found by: Chasing the octyl-palmitate dilution series back to the paper it was attributed to, and finding the paper contains no such table.

How a claim gets killed

Every page here goes through an adversarial external read before it publishes. It is handed to a model with the primary documents attached and instructions to assume the page is wrong until it has checked. The build refuses to publish a page that has not been through it.

The rule underneath all of it is one sentence: we do not publish a score we have not read in its own document. Nearly every number in circulation fails that test.

‹ Check an ingredient list · How we decide