One in Three Peptides Fails Testing? We Chased the Number to Its Source (2026)
The 33%, 42% and 71% failure figures all come from one dataset. What they measure, what a certificate of analysis settles, and what it never covers.
Important notice: This article discusses published analytical data and press reporting for scientific information only. It is about labelling accuracy and batch documentation, not about what any peptide does in a body. We sell research peptides, so we have an obvious interest in this topic: none of the studies discussed here tested our products, and we say so rather than implying otherwise.
TL;DR: three numbers, one dataset
The claim: about one in three peptide products fails basic quality checks. It has been circulating since a Guardian report in April 2026 and was pushed back into circulation by an advocacy group in July. The source: a single testing and ratings platform in Texas. The Guardian calls it a laboratory; its own methodology page says it sends samples to third-party labs. The harsher numbers: 41.6 percent and 71.1 percent come from a re-analysis of that same platform's public dataset, published as a preprint that states on its face that it is not peer-reviewed. The critical detail: the spread between 33, 42 and 71 percent is almost entirely about where the pass line is drawn, not three independent measurements. What actually fails: not purity. Median purity in the re-analysis was 99.80 percent. The dominant failure mode is dosing accuracy, the milligrams in the vial versus the milligrams on the label. What no certificate covers unless ordered separately: sterility, endotoxin, heavy metals, residual solvents, particulates and stability.
Where the number comes from
On 6 April 2026 the Guardian published a report on the peptide market by Sarah Marsh and Nicola Davis. In it, Finnrick, described by the paper as "a peptide testing laboratory in Texas", said that "about a third of the thousands of products it analysed failed basic quality checks, and that proportion had stayed broadly unchanged over the 12 to 14 months it had been gathering data".
That single sentence is the origin of every "one in three" headline you have seen since.
The Guardian also reported the three failure categories: identity, meaning the substance is not what the label claims; purity, "with anything below the 98% threshold considered substandard"; and quantity, where the vial contains more or less than the stated dose. And it reported something that matters a great deal for how much weight the figure carries: that Finnrick "typically buys peptide products itself for testing, saying it prefers to analyse samples bought through the same channels used by ordinary customers rather than accepting products directly from vendors". The same paragraph adds that "in June last year, the lab also began accepting samples submitted by members of the public", and that "any products that came directly from vendors were flagged as such on its website", so the model the Guardian describes is a mixed one, not purchases only.
Three things that do not line up
We went to the source rather than repeating the quote, and found three discrepancies worth knowing. It does not call itself a laboratory. Finnrick's methodology page states that it "sources samples from vendors and sends them to third-party labs", and it maintains a separate page listing partner testing labs. "Laboratory" is the Guardian's word. The 98 percent threshold is not Finnrick's. On its live methodology page, every scoring mode uses a purity cutoff of 99.5 or 99.8 percent. The figure 98.0 percent does appear elsewhere in this story: it is the purity threshold of the looser of the two models in the re-analysis discussed below. Its own pages disagree about where the samples come from. The methodology page says samples are sourced from vendors. The company page says "Samples come from public submissions and Finnrick purchases." The homepage, fetched on 24 July 2026, says "Customers sent us 9,041 peptide vials." The Guardian's fuller account, own purchases plus public submissions since June of the previous year with vendor-supplied products flagged, matches the company page; the discrepancy that remains is between Finnrick's own pages. It matters because a dataset built mainly from purchases the platform makes itself, like an ordinary customer, supports a claim about the market far better than one built mainly from what people choose to send in.
The 71 percent version
In July 2026 a second set of numbers began circulating: 71 percent, 42 percent, 15 percent. These come from a re-analysis of Finnrick's public dataset by two researchers at a private institute, published on a preprint server. The page carries the line "This version is not peer-reviewed."
Its headline finding, in its own words: "Between the two models, 41.6% to 71.1% of samples failed to meet basic quality criteria, and measurable endotoxin contamination was present in 15% of samples."
Both models are the authors' own construction. One approximates compounding-pharmacy expectations, requiring measured abundance between 90 and 110 percent of label and purity of at least 98.0 percent. The other applies tighter manufactured-product criteria. The authors state plainly that neither model represents a formal specification for any individual peptide, and that they are benchmarks for discussion.
Why 33, 42 and 71 percent are not three findings
This is the single most important thing to understand about all of these numbers. They come from one dataset: the samples that reached one platform. What differs between them is where the pass line sits. Move the abundance window from a loose tolerance to 90 to 110 percent of label and the failure rate rises by roughly a quarter, from about a third to 41.6 percent. Move it to manufactured-product tolerances and it rises by roughly seventy percent again, to 71.1 percent. Across the whole span that is 71.1 against about 33, a single doubling produced by the pass line. A reader who sees three numbers and concludes that three studies found three failure rates has been misled by arithmetic, not by data.
What actually fails, and it is not what people assume
Here is the finding that changed how we read this whole story. In the re-analysis, the median purity across the dataset was 99.80 percent, with an interquartile range of 99.50 to 99.90 percent. The median measured abundance was 101.80 percent of label claim, with an interquartile range of 95.00 to 109.00 percent.
So the typical vial is both pure and filled close to its label. What the strict models expose is the spread around that median: once a 90 to 110 percent window is enforced, dosing accuracy is where the failures concentrate, and the authors call it "the dominant quality deficit across the dataset as a whole".
Under the stricter of the two models, 44.3 percent of samples met the purity criterion but fell outside the abundance window. For tirzepatide the pattern was even sharper: only 13.8 percent passed both criteria, while 58.8 percent met purity and failed abundance, which the authors describe as consistent purity paired with frequent dosing inaccuracy.
Purity and content are different measurements
Purity is a ratio: the share of everything the instrument detects that is the target peptide. Content, sometimes called abundance or assay, is a mass: how many milligrams are actually in the vial. A vial that has lost half its fill can still test 99 percent pure on what remains, because removing peptide does not change the ratio. This is why a certificate that reports only a purity percentage has answered the easier half of the question.
Identity failures, meaning the vial did not contain the stated peptide at all, ran at 2.4 percent of samples overall. The re-analysis reports identity-failure rates for only four of its fourteen peptides, and names them by rank rather than as a range: TB-500 highest at 10.0 percent, CJC-1295 second at 9.1 percent, tesamorelin 3.6 percent and BPC-157 3.5 percent.
The caveats the authors themselves published
A rare thing about this preprint is that its limitations section is more useful than its headline. Two passages deserve quoting.
On whether the sample is representative:
"First, the Finnrick Analytics dataset is not a random or representative sample of the gray market peptide supply. Samples are submitted voluntarily by consumers and vendors, which introduces selection bias. Companies confident in their product quality may be more likely to submit samples, and poor-performing batches may be systematically underrepresented."
On what was not measured at all:
"The absence of data on sterility, residual solvents, particulate matter, stability, and structural integrity means that the failure rates reported here should be understood as a lower bound on the true quality deficit in this supply chain."
Those two statements point in opposite directions, which is exactly why the honest answer to "what is the real failure rate?" is that nobody knows. The endotoxin figure has a further limit the headlines dropped: endotoxin data existed for 243 samples out of 6,487 screened, under four percent, so the 15 percent contamination figure describes those 243 and not the dataset.
Who put the number back into circulation, and why that matters
The July 2026 revival came from the Partnership for Safe Medicines, in a post that links the phrase "a third of the peptide samples they test fail identity, purity, or quantity checks" directly to the Guardian article.
The organisation describes itself as a public health group comprising more than 45 non-profit organisations. Its own board page names its board president as the former president and chief executive of the Pharmaceutical Security Institute, the pharmaceutical industry's anti-counterfeiting body. Kaiser Health News reported in 2017 that the group had been led for a decade by a senior vice president of PhRMA, the pharmaceutical manufacturers' trade association, who served as its principal officer until that year, and that the group describes PhRMA as a dues-paying member. It is registered as a 501(c)(6) business league rather than a public charity, and it does not publish its funders.
How to weigh that
An interested source can still be right, and this one is pointing at a real dataset rather than making something up. The correct adjustment is not to dismiss the figure but to notice what it is being used for. A trade-adjacent group campaigning for tighter restrictions on compounded and grey-market peptides has an obvious reason to circulate the highest defensible failure rate. We sell peptides, so we have an obvious reason to circulate the lowest one. Neither of those incentives changes what the underlying dataset measured, which is why this article is about the measurement.
What the peer-reviewed literature actually contains
Set the platform data aside and ask what has been published under peer review about what is in grey-market peptide vials. The answer is: not much, and mostly not about the compounds people are buying today. One entry below is not a peptide study at all: selective androgen receptor modulators are small molecules, and that row is listed as a comparator for what online grey-market sellers ship, not as evidence about peptides.
- What was tested
- 44 products bought online
- Finding
- Only 52 percent contained the advertised compound class. 39 percent contained a different unapproved drug. In 9 percent no active compound was found at all.
- What was tested
- Vials from 3 online shops
- Finding
- 4.32 to 8.84 mg per vial against a 10 mg label, so 43 to 88 percent of the claimed content. Two shops' vials carried unknown impurities of 4.1 to 5.9 percent.
- What was tested
- 3 vials bought without prescription from illegal online pharmacies
- Finding
- Content exceeded the label by 28.56 to 38.69 percent, measured purity was 7.7 to 14.37 percent against the 99 percent claimed on the labels, and endotoxin was detected in all 3 vials.
- What was tested
- Customs seizures
- Finding
- Glycine-modified analogues of GHRP-2, GHRP-6 and ipamorelin, and a growth hormone variant carrying an extra amino acid.
Notice what is missing: there is no peer-reviewed, large-sample analysis of the compounds that dominate demand right now, such as BPC-157, TB-500, retatrutide or CJC-1295. The only large dataset covering those is the unreviewed preprint. Anyone who tells you the failure rate for modern research peptides is scientifically established is overstating what exists.
What a certificate of analysis settles, and what it does not
Identity: what is it?
Normally mass spectrometry, comparing the measured molecular weight against the expected one. This is the test that catches the 2.4 percent case where the vial contains something else entirely, and it is the one a purity percentage cannot substitute for.
Purity: how homogeneous is it?
Normally HPLC, expressing the target peptide's peak area as a share of the total detected. It tells you about truncated or oxidised versions of the same peptide. It tells you nothing about how much is in the vial.
Content or assay: how much is in there?
A mass, in milligrams per vial. On the dataset above this is where most failures actually sit, which makes it the number worth checking first, against both the label and the batch's own report.
Everything else: only if separately ordered
Sterility, bacterial endotoxin, heavy metals, residual solvents, particulates and stability are distinct tests with distinct prices. A standard identity-and-purity certificate does not include them, and their absence from a report is not a pass, it is silence.
The limit of verification
Serious laboratories let anyone confirm that a report is genuine. Janoshik, which issues most of our suppliers' certificates, runs a public lookup requiring a task number plus a unique key, so a report can be checked on the lab's own server rather than taken on trust from a PDF. Reports issued by other laboratories are verified through that laboratory's own system, not through Janoshik's. That closes one gap and leaves another open: verification proves the report is authentic and describes the batch it says it describes. It cannot prove that the vial in your hand came from that batch. No certificate anywhere solves that, which is why batch numbers on labels and per-batch publication matter more than the purity figure everyone quotes.
Our own position, stated plainly
We publish the batch reports we hold, from our suppliers' third-party laboratories, on our CoA page. Those reports cover identity, purity and content, and, as described above, not sterility, endotoxin or heavy metals unless a specific panel was commissioned. We do not run our own laboratory and we do not test our own products.
None of the datasets in this article tested anything we sell, and we are not going to pretend otherwise by putting our catalogue next to someone else's failure rate. What we can say is what the measurement means, so that a reader can judge any supplier, including us, on the same terms: is there a batch report, does it cover identity as well as purity, does it state the measured content, and can it be checked against the issuing laboratory.
Further reading
Our guide to vetting a research peptide supplier walks through reading a certificate line by line, and how to spot a fake CoA covers document forgery specifically.
Related pages
Bacteriostatic water and research supplies
Compounds named in the identity-failure data
Full-length 43-amino-acid Thymosin Beta-4, a naturally occurring repair protein, independently confirmed by a third-party CoA from Janoshik. Promotes cell migration and new blood vessel formation for systemic tissue healing. Especially researched for muscle, tendon, and cardiac repair.
CJC-1295 without DAC (Mod GRF 1-29) is a short-acting GHRH(1-29) analog for GH/IGF-1 research. Research-grade lyophilized powder, specified purity >=99% (HPLC). Laboratory use only.
Gastric pentadecapeptide (15 amino acids) known for exceptional tissue repair properties. Promotes wound healing, angiogenesis, and cytoprotection across tendons, muscles, gut, and nerves. Over 30 years of preclinical research.
Handling and storage
USP-grade sterile water with 0.9% benzyl alcohol (near-neutral, ~pH 6) - the standard solvent for reconstituting lyophilized peptides. Essential accessory for any peptide research. Each vial is sealed and ready to use.
Transparent storage box with 10 individual compartments for 1-3 ml peptide vials. Stackable, fridge-friendly, travel-safe. Ideal for organising bacteriostatic water, GLP-1, BPC-157, and similar vials.
FAQ
Sources
-
Marsh S, Davis N. "Traceability is vital: labs test thousands of unregulated substances amid peptide craze." The Guardian, 6 April 2026. https://www.theguardian.com/science/2026/apr/06/labs-testing-thousands-of-unregulated-substances-amid-peptide-craze
-
Mendias CL, Awan TM. "Evaluation of Research Grade Peptides Marketed Directly to Consumers Reveals Extensive Variability in Purity and Measured Abundance." Preprints.org, manuscript 202604.1748. Not peer-reviewed. https://www.preprints.org/manuscript/202604.1748
-
Partnership for Safe Medicines. "The five risks of taking unapproved peptides." Published 13 July 2026. https://www.safemedicines.org/2026/07/five-reasons-bulk-peptides.html
-
Lupkin S. "Nonprofit Linked To PhRMA Rolls Out Campaign To Block Drug Imports." Kaiser Health News, 19 April 2017. https://khn.org/news/non-profit-linked-to-phrma-rolls-out-campaign-to-block-drug-imports/
-
Van Wagoner RM, Eichner A, Bhasin S, Deuster PA, Eichner D. "Chemical Composition and Labeling of Substances Marketed as Selective Androgen Receptor Modulators and Sold via the Internet." JAMA 2017;318(20):2004-2010. PMID 29183075. https://pubmed.ncbi.nlm.nih.gov/29183075/
-
Breindahl T, Evans-Brown M, Hindersson P, et al. "Identification and characterization by LC-UV-MS/MS of melanotan II skin-tanning products sold illegally on the Internet." Drug Testing and Analysis 2015;7(2):164-172. PMID 24771717. https://pubmed.ncbi.nlm.nih.gov/24771717/
-
Ashraf AR, Mackey TK, Vida RG, et al. "Multifactor Quality and Safety Analysis of Semaglutide Products Sold by Online Sellers Without a Prescription." Journal of Medical Internet Research 2024;26:e65440. PMID 39509151. https://pubmed.ncbi.nlm.nih.gov/39509151/
-
Gajda PM, Holm NB, Hoej LJ, et al. "Glycine-modified growth hormone secretagogues identified in seized doping material." Drug Testing and Analysis 2019;11(2):350-354. PMID 30136411. https://pubmed.ncbi.nlm.nih.gov/30136411/
-
Krug O, Thomas A, Malerod-Fjeld H, et al. "Analysis of new growth promoting black market products." Growth Hormone & IGF Research 2018;41:1-6. PMID 29864719. https://pubmed.ncbi.nlm.nih.gov/29864719/
Research disclaimer: All content serves scientific information only and concerns labelling accuracy and analytical documentation, not the biological effects of any compound. Our products are supplied for laboratory research use only and are not intended for human consumption.
Research context for English-speaking buyers
Most of our English-speaking customers ship to the UK, Ireland, Malta or other English-as-second-language EU territories. The regulatory picture differs per country.
- Relevant authorities
- MHRA (UK, post-Brexit), HPRA (Ireland, EU-aligned), FDA Section 503A bulks list (US, restricted Cat 2 status of several peptides as of 2026)
- Customs and VAT
- EU shipments include 19% VAT; UK shipments after Brexit are now extra-EU and may attract UK VAT plus a handling fee at import
- Typical shipping window
- EU 2-4 working days, UK 4-7 working days, other international 7-14 working days, depending on customs
Research-grade peptides shipped from our EU warehouse are sold for laboratory use only and are not authorised for human or veterinary therapeutic application in any of the destination jurisdictions. US customers should be aware that the FDA Section 503A bulks list classification (and the April 2026 reclassification of twelve compounds) only governs compounding pharmacies, not direct-to-researcher imports for non-clinical work. UK buyers should declare the consignment on import and may be asked for a research justification by HMRC. We provide a CoA per batch identified by colour code rather than serial number; customs sometimes asks for this document when clearing the parcel.