Corrections · and what we do not cover
So this page carries two things almost nobody publishes: every finding we have withdrawn, and a plain list of what the audit does not cover. Both admit fallibility, which is the point. A vendor who names their own limits is the one worth believing about everything else.
Corrections record
Every finding withdrawn or materially amended, whether a brand asked us to look again or our own review caught it first. Nothing is added for effect and nothing is removed once added.
Two tools computed the same published totals different ways, and the difference only became visible when we published a category whose listings overlap ones already scanned. Our figure script counted a finding once per ruling while counting a listing once overall, so a product selling into two categories contributed its findings twice to a total printed beside a listing count that counted it once. The deploy check, which deduplicates, disagreed by 116 findings. A related fault gave the phrase-frequency table one to three extra brands on seven of its ten rows.
Outcome. Both tools now compute every published figure the same way, from the same deduplicated basis, and agree exactly. The energy category was published the same day: listings ruled 1,396 to 1,535, brands 550 to 595, findings 2,413 to 2,624, and brands carrying a Critical 112 to 119. The Critical share held at 20%. Every phrase-frequency count was restated. The brand count went out at 636 for about an hour before being restated to 595: the new category had arrived without brand names, because Amazon's search card does not carry one, so our counter had been guessing company names from product titles and counted one company as several. The names were recovered from the pages we had already cached. The figures above the change were correct for the seven categories they covered. What was wrong was that our two ways of counting would have diverged the moment an overlapping category was added, and it took adding one to find out.
Two faults in how we identified brands, both in the sports category. Our counter accepted "Sponsored", which is Amazon's advertising label rather than a company name, as a brand: listings from as many as 37 different companies were counted as one. Separately, that category had been ingested without brand names at all, so the counter fell back to guessing a brand from each product title, which splits one company across several spellings.
Outcome. Both corrected the same day. The counter now rejects "Sponsored", and the real brand names were recovered from the detail pages cached at scan time, so every sports listing now carries the brand Amazon publishes. Brands ruled restated from 548 to 550, brands carrying a Critical from 114 to 112, and the Critical share from 21% to 20%. Five phrase-frequency counts moved by one to five brands each. The figure moved twice within the hour as the fix went deeper, and 550 is the count from published bylines rather than inference. No finding, verdict, or listing count changed, and no client report was affected.
The engine flagged "Studies show that the combination of alpha-glycerylphosphorylcholine" on a cognitive listing as an uncited study claim. Review found the surrounding copy did cite the study, and the finding was a compliant reframe rather than a violation.
Outcome. Finding withdrawn. It was never sent to the brand.
The engine returned two Critical findings on "helps relieve stress" and "help to relieve stress", reading them as symptom-relief claims. Stress is not a disease and that phrasing is ordinary structure/function language. The engine's own note hedged on both.
Outcome. Both withdrawn and the brand removed from the send list. A message calling that a Critical would have been wrong.
A warning-letter target was recorded as still publishing the exact phrase the FDA had quoted. Checking the live site showed the phrase had been removed and only the product name remained.
Outcome. Record corrected. What holds is that the letter is public and permanent, not that the claim is still up.
Every entry above marked “Before issue” was caught by our internal review before the finding reached a brand. That is not luck. Machine output needs checking, and our internal review is there to check it. On client files, a licensed pharmacist reviews every finding before signing. When one gets through anyway, it appears here with the same detail.
Report a finding
Anyone can report one. You do not have to be a client. Send the finding and why you think it is wrong to audit@claimsverified.org.
We answer within five business days with one of three outcomes: the finding stands and here is why, the finding is amended, or the finding is withdrawn.
Withdrawn or amended, it goes on this page. Named only if you want it named. The entry goes up either way.
The limits
Every line here is a question a buyer eventually asks. Better answered before you pay than after.
Nothing we do says anything about what is in the bottle: not purity, not potency, not dose, not contamination, not whether it works. USP and NSF certify that. We certify what is said about it.
We do not review manufacturing, GMP compliance, or any supplier. Where a listing claims a facility credential, we check whether the claim is supportable from the listing, not whether the facility holds it.
A finding is a professional opinion under published standards, signed by a licensed pharmacist. It is not a legal opinion and we are not your lawyers. Where a brand needs counsel we say so and refer up.
No regulator reviews or approves a supplement claim before it is published, and nothing we issue changes that. A pass from us is not a defence, it is evidence of the review you did.
The audit covers the assets listed in the report. Copy changes constantly, so a report is a snapshot with a date on it, which is why monitoring exists.
The engine produces findings and a pharmacist reviews them, and that review exists because machine output needs checking. When something gets through anyway, it goes in the record above.
What is left after all of that is one thing done properly: every claim you publish, ruled against 109 published standards, with the rule each one runs into and a rewrite at the strongest wording the standard allows, signed by a licensed pharmacist.
See where your own listing sits. One Amazon listing or your website, checked free. No call, no account.
Check my listing