Verified Beauty Data

Corrections / The record

What we got wrong

A site that claims every number is checkable has to show what happened when a number wasn't. This log is append-only: entries are never edited or removed, and if an entry turns out to be wrong itself, it gets a new entry rather than a quiet rewrite. Errors caught before they shipped are not listed here — that is the method working, not a correction.

Found something wrong? hello@verifiedbeautydata.com

Log / Newest first

6 corrections

  1. LED mask FDA study (/research/led-face-mask-fda-check/), its CSV/JSON and charts, and the LED mask buyer's guide counts: 15 of 74 masks. Of the 24 listings that say "FDA cleared," "tied to that mask" went from 6 to 9 and "no FDA record found" from 8 to 4

    LED mask FDA study: 15 masks were in the wrong group, and two headline numbers were wrong

    What was wrong
    The study said only 6 of 24 Amazon LED masks that say "FDA cleared" had a filing tied to that mask, and that 8 had no FDA record at all. Four of those 8 (MICEYLIVE, Vital Vibes, Gleelar, Aureskyn) show a model number on Amazon that appears in a factory's FDA filing. Three masks we called "not tied" are tied: FDA's device listing names RENPHO's mask and model, NOMADARA gives a K number whose filing lists its model, and Nanoleaf's own site ties its model to a K number whose FDA listing matches Amazon's part number. The method also said a matching model number alone made a mask "tied," which is not the rule the data used.
    Why it happened
    Our first pass ran targeted FDA searches and did not use Amazon's part-number field or every FDA device listing. The method text was written loosely and did not match how masks were sorted.
    What changed
    We downloaded FDA's complete records for LED masks (every OHS, OLP and ILY clearance, the text of 397 filing PDFs and 962 device listings), read every listing photo of the 24 "FDA cleared" masks, and searched all of it for every mask. The rule is now written out: a mask is tied when an FDA record links the brand to it; a model match alone is "factory filing, brand not named." Each changed row in the data is marked, with its earlier finding and the evidence.
    Found by
    Our own re-check while answering a reporter's questions about the method
  2. LED mask buyer's guide (/red-light-therapy-masks/): 3 of 74 masks; the "say FDA with no record" count fell from 11 to 8

    Three LED masks were listed as "says FDA, no record found" when that was wrong or unfair

    What was wrong
    The guide put the iRestore LED Face Mask, the EUARNE mask and the SDKWDH mask in the group that says "FDA" with no FDA record found. iRestore's face mask listing and product page do not say FDA (its FDA clearances are for its hair devices). EUARNE's listing cites K250413, a factory's clearance for its LED masks. SDKWDH's model name, M01, appears in three companies' FDA filings.
    Why it happened
    iRestore was placed on a third-party description of the mask, not the brand's own words. EUARNE and SDKWDH were held to a stricter rule than other masks whose listings point to a factory filing we could not confirm.
    What changed
    iRestore is now "no FDA claim, no record found." EUARNE and SDKWDH are now "likely filing, not tied to this mask," the same group as every other mask in that position. All 74 listings were re-read and every "says FDA" mask was searched at FDA again on October 13, 2026.
    Found by
    Our own re-check before publishing the LED mask FDA study
  3. Homepage claim; 114 product prices

    Prices were described as weekly-checked when they were not

    What was wrong
    The homepage said every price is re-checked against Amazon weekly. In fact the most recent price sweep across the catalogue ran on 2026-08-20, and three products carried prices last verified on 2026-07-30 — 18 and 39 days old respectively at the time this was caught.
    Why it happened
    The claim was written when the price sweep was running frequently, and was never revised when the cadence slipped. Each individual product page did carry its true "price as of" date, so the per-page numbers were honest; the site-level summary of them was not.
    What changed
    The homepage claim now describes the actual cadence rather than an aspirational one, and points to the per-product verification date that was always there.
    Found by
    Our own audit
  4. 3 pages

    Three dupe reports displayed "last checked undefined"

    What was wrong
    On the SkinCeuticals C E Ferulic, SkinCeuticals Phloretin CF and Crème de la Mer dupe reports, the note explaining that a candidate product had left Amazon read "last checked undefined" instead of naming a date.
    Why it happened
    The template read a date field that is only filled in for products the automated liveness gate marks as not buyable. For products delisted before that gate existed, the field was empty and rendered as the literal word "undefined".
    What changed
    The template now falls back to the date the product record was last verified, and renders no parenthetical at all when no date is known rather than printing a placeholder.
    Found by
    Our own audit
  5. 24 pages

    24 ingredient pages promised a ranking that was never coming

    What was wrong
    Ingredient pages for allantoin, ceramides, centella, panthenol, urea and 19 other actives carried a section headed "measured product rankings coming soon", above a note about ranking by price per gram of active, above no ranking at all.
    Why it happened
    The heading treated a real result as an unfinished task. There was no ranking on those pages for a substantive reason: a price per gram of active requires a disclosed percentage, and not one product we had checked published one for those ingredients. Calling that "coming soon" implied the work was pending when in fact the answer was already in.
    What changed
    Those sections now state the finding — that no brand publishes a dose for the ingredient, and why that makes any ranking dishonest — and then list the products confirmed to contain it at an unstated amount. Nothing was removed; the framing was wrong, not the content.
    Found by
    Our own audit
  6. 27 links across 21 products — 11.8% of all buy links

    27 buy links pointed at Amazon listings that no longer existed

    What was wrong
    Buy buttons on product and dupe pages linked to Amazon listings that had been delisted. A reader clicking one landed on a listing they could not buy from, and six of those products were already marked as not buyable on our own pages while still showing a live buy button.
    Why it happened
    The internal stock flag only drove the availability field in structured data. Nothing in the page templates actually suppressed the link, so a product could be marked as not buyable and still render a working buy button.
    What changed
    The stock flag now suppresses the buy path everywhere it appears — the buy-the-original call to action, both verdict-card lanes, and the scoreboard row. Four links were repointed to a live listing of the identical size; the other seventeen were flagged rather than swapped, because every replacement the API offered was a different strength, formula or size, and silently substituting one would have falsified the price-per-gram figure that is the whole point of the ranking.
    Found by
    Our own automated listing-liveness gate

Policy / How this works

Our corrections policy

If a formula here doesn't match the bottle in your hand, or a study doesn't say what we claim it says, that is a verifiable error. Tell us what the label or the paper actually says and we will re-verify it against the primary source. If we were wrong, the page is corrected, its Verified date is updated, and the error is logged here with what caused it.

We do not silently edit. Substantive factual corrections get an entry on this page. Typos, formatting and layout changes do not — nobody needs a changelog of comma fixes.

The method that is supposed to prevent these errors is documented in full on the about page, and who is accountable for it is on the author page.