Corrections log

30 August 2026 · Medical disclaimer

What was wrong. The page said that on 31 March 2026 FDA issued seven warning letters stating that despite labeling products for laboratory research purposes only, website evidence established the products were intended to be drugs for human use, and it listed all seven numbers behind that sentence. Five of the seven say that. Two do not: 722215 and 722228 carry no research use language at all, because those products were offered to consumers outright rather than under a research label. The sentence attributed a finding to two letters that do not contain it.

What it says now. That all seven rest on intended use established by evidence obtained from the seller's website, that five of them make the point about the research use label itself and are named, and that the other two reach the same finding without that language and are named too. The legal reasoning the section is making is unchanged, because it was never the research label that decided these cases.

How it was found. A review of the question explainer that reused this wording, which read each of the seven letters rather than the sentence they had been summarised into. The explainer was corrected in the same change, which is why both surfaces now say the same thing.

30 August 2026 · GHK-Cu, melanotan II, MOTS-c, and the group and comparison pages that carry them

What was wrong. Eight ClinicalTrials.gov records share one industry sponsor, and three of the eight describe themselves in their own registered text as an example record, a mock study or a fictional study. The TB-500 and BPC-157 reviews already read that sponsor account in full, though the BPC-157 review summarised those three self-descriptions as saying the records "are not studies", which is one step past what two of the three say: an example record and a mock protocol call themselves that, not not-a-study. Eight other pages did not read the account at all, and presented its records as ordinary trials. The GHK-Cu review went furthest: it called one of them "a live, registered, vehicle-controlled trial" and "the first registered study capable of changing" the human picture, and said the record on the page was "likely to look different within about eighteen months". The sexual-health group page, PT-141 vs melanotan II and melanotan I vs melanotan II counted the melanotan II record as a recruiting trial when its own registered summary opens "This example interventional study record describes". The longevity group page and MOTS-c vs humanin counted the MOTS-c record as the one registration that administers the peptide, which is what the record proposes rather than something it has done. The healing and recovery group page and GHK-Cu topical vs injected carried the GHK-Cu record with no sponsor account beside it.

What it says now. Every one of those pages states what its record contains and states the sponsor pattern as fact beside it. The three records that describe themselves that way are quoted doing it, on the BPC-157 review as well as everywhere else. The other five are described by what they contain, never as self-declared, each with the account they sit in named, and no page states a forward-looking expectation from any of them or counts a trial as underway. The GHK-Cu review's grade is unchanged: it comes from two completed trials, in 1992 and 2006, and never rested on a registration. Six pages gain a sponsor-search reference appended to the end of their lists; healing and recovery and GHK-Cu topical vs injected amend the registry reference they already carried. Every source count and category split those pages state was recomputed from the reference arrays rather than adjusted by hand, which turned up four numbers that were already wrong before this change: the healing and recovery page's total and its primary-paper count, and the sexual health page's total and its primary-paper count.

How it was found. Pulling all eight records from the registry API one at a time and classifying each one on its own text, rather than assuming the pattern held uniformly, then grepping the whole corpus for every identifier and for the sponsor name. The three self-declarations are not identical and two of the eight reproduce a large manufacturer's own protocol identifiers for that manufacturer's own molecules, so a per-record read was the only way to say the right thing about each.

23 August 2026 · Dihexa, Fosgonimeton, and Follistatin-344

What was wrong. Four references carried names of people who are not authors of the cited paper, and a fifth showed the wrong first author. On the dihexa and fosgonimeton pages, a 2018 systematic review of angiotensin IV cognition studies (PMID 29733881) was shown with authors from the laboratory that originated the compound series, when the review's authors are Ho and Nation, who are independent of that laboratory. On the dihexa page, a 2021 mouse study (PMID 34827486) was shown with three names that appear nowhere on the paper, and a stem-cell protocol (PMID 27532814) carried a name from the adjacent reference and the wrong first author. On the follistatin-344 page, a 1998 review (PMID 9785474) was attributed to two authors of a different follistatin review published the same year. On the IGF-1 LR3 page, a fetal sheep study (PMID 33427051) listed the right group under the wrong first author.

What it says now. Each reference carries the author list from its PubMed record, in full. No claim on any of these pages rested on the wrong names, but this site reports on authorship concentration as a quality signal, and a reference list that misassigns an independent review to the originating group is wrong in exactly the dimension we ask readers to trust.

How it was found. Running the corpus-wide reference audit, which fetches every stored reference's live PubMed record and compares it field by field. The same pass replaced every hand-shortened author list with the record's full list and moved editorial annotations out of title fields, so every PubMed-backed title is a verbatim transcription.

23 August 2026 · Dihexa

What was wrong. The page said the compound is "not a compound any regulator has evaluated for human use", and its regulatory section opened "There is no agency position on this compound because no agency has been asked to take one." FDA lists dihexa acetate among bulk drug substances for use in compounding that may present significant safety risks, and publishes its reasoning. The page had been written without checking that list.

What it says now. That there is one agency position and it is a statement about missing information, quoting FDA: it has identified no human exposure data on drug products containing dihexa acetate by any route, and lacks important information regarding any safety issues raised by it, including whether it would cause harm if administered to humans. The listing is now a dated row in the regulatory table with the document behind it, and the page links to the 503A record.

How it was found. Reading FDA's category 2 page while building the CJC-1295 pages, which are on the same list, and noticing dihexa in the same table.

23 August 2026 · Argireline vs SNAP-8

What was wrong. The page read: "Acetyl hexapeptide-8 has one randomised, placebo-controlled human trial." That described the published wrinkle literature and presented it as the whole record. ClinicalTrials.gov holds six registered trials of the ingredient, two of them testing the marketed mechanism against a scored clinical endpoint with results posted to the registry.

What it says now. That the published wrinkle result is one randomised placebo-controlled trial, and that the registered record is larger than that trial suggests, at six trials including two at a clinical endpoint. All six are set out on the compound page.

How it was found. Running a registry sweep before building the compound page, rather than taking the published literature as the record.

23 August 2026 · Immune peptides group page

What was wrong. The page said thymosin alpha-1 carries 65 registrations, 33 of them phase 2, 8 phase 3 and 10 phase 4. Those figures counted what a registry query returned. A free-text search for a thymosin name also returns thymosin beta-4 records, which are a different molecule with its own page on this site, so the count included trials of something else.

What it says now. 62 registrations name thymosin alpha-1 in their interventions, of which 32 are phase 2, 9 are phase 3 and 7 are phase 4, from five queries returning 91 distinct records. The method is stated alongside the number, because a different filter gives a different count.

How it was found. Re-running the sweep with an intervention filter while building the thymosin alpha-1 page, and comparing the result with what the group page already said.

23 August 2026 · Growth hormone secretagogues group page, Sermorelin vs CJC-1295, and CJC-1295 vs CJC-1295 with DAC

What was wrong. Three sentences across those pages attributed the single registered CJC-1295 trial to the version carrying the albumin-binding linker. The registry record names its intervention only as "CJC 1295", carries no intervention description and lists no primary outcome, so it cannot be assigned to either product sold under that name.

What it says now. That one registration exists under the CJC-1295 name, that it is recorded as terminated with no stated reason, and that which version it used is not established.

How it was found. Reading the registry record field by field while building the two CJC-1295 pages. The research note written before that build listed this as a thing not to assert, which is what made the existing sentences visible.

23 August 2026 · Immune peptides group page, Thymosin alpha-1 vs thymulin, and Thymosin alpha-1 vs thymosin beta-4

What was wrong. Four places stated that thymosin alpha-1 is approved as a medicine in more than thirty countries. That figure had been on the site since 2 August 2026 with no primary source behind it, and no regulator's register was checked for it.

What it says now. That it is registered as a medicine outside the United States, which the record supports. The number survives in one place, on the compound page, labelled as not verified against any regulator's register.

How it was found. Trying to source it while building the compound page, and not being able to.

8 August 2026 · homepage, the evidence gap section

What was wrong. The section read: "Everything sold today rests on animal studies and community reports." That overstated the position. Published human studies of BPC-157 do exist outside the two registrations, in groups of roughly a dozen to two dozen people, and FDA described several of them to its Pharmacy Compounding Advisory Committee in July 2026.

What it says now. That the small human studies exist, that FDA told the committee they were not enough to reach a conclusion on effectiveness, and that the record past them is animal studies and community reports. The headline figure is unchanged: there is still no completed, published human efficacy trial on the registries.

How it was found. Reading FDA's own meeting documents for that committee, which are summarised on the 503A record.