Data
6,285 Peptide Test Reports: What the Largest Public Dataset Shows
Most writing about peptide testing is advice: check the lab, look for a lot number, be wary of 99.9%. A preprint posted in April 2026 does something different. It counts. Two researchers took a public dataset of third-party test reports and asked how often research-grade peptides actually matched their labels. The answer depends heavily on the standard applied, and that dependence is the interesting part.
What the dataset is
The authors, Christopher L. Mendias and Tariq M. Awan, analyzed a deidentified dataset from Finnrick Analytics, a third-party testing company (source 1). Reports ran from December 17, 2024 to April 3, 2026 and covered 14 peptides sold by 203 vendors (the full list is in the preprint, source 1).
The paper's own sample counts do not fully agree. The abstract reports 6,441 samples. The Methods section reports 6,487 analysis reports, of which 156 lacked the stated peptide and 46 had incomplete data, leaving 6,285 analyzed (96.9%). We use 6,285.
What was measured
- Purity by HPLC with UV or mass-spectrometry detection, as a percentage.
- Abundance, meaning the measured amount of peptide expressed as a percentage of the label claim.
- Endotoxin, by LAL or recombinant assay, in only 243 reports (under 4% of the dataset).
Purity and abundance answer different questions. Purity is a ratio of peak areas. Abundance asks whether the vial contains the amount the label says. Our articles on purity vs. content and HPLC vs. LC-MS explain why one does not stand in for the other. This dataset shows the gap at scale.
Email sign-ups are paused.
Two yardsticks, two very different results
The authors applied two acceptance models. They label these "compounded" and "manufactured" standards, and say they are not formally codified specifications or regulatory determinations (source 1).
| Measure | Compounded-standards model | Manufactured-standards model |
|---|---|---|
| Abundance (% of label claim) | 90 to 110% | 95 to 105% |
| Purity | at least 98.0% | at least 99.5% |
| Samples meeting both | 58.4% | 28.9% |
| Samples failing | 41.6% | 71.1% |
That spread is the paper's headline: between 41.6% and 71.1% of samples failed to meet basic quality criteria, depending on the model. The medians look reassuring on their own. Overall median abundance was 101.80% of label claim (interquartile range 95.00 to 109.00%), and median purity was 99.80% (99.50 to 99.90%). Most of the failures come from the tails of the distribution and from how strict the cutoff is, not from the middle.
Where compounds differed
The authors report large differences between peptides. A few examples from the paper:
| Peptide | What the paper reports |
|---|---|
| GHK-Cu | Highest median purity, 99.90% |
| TB-500 | Lowest median purity, 97.29%; identity failures in 10.0% of samples |
| CJC-1295 | Median purity 98.49%, with a lower quartile of 91.59%; identity failures in 9.1% |
| Tesamorelin and BPC-157 | Identity failures in 3.6% and 3.5% |
We quote "identity failure" as the authors report it. We did not read how they define it, so check the paper before leaning on those percentages.
Purity did not predict endotoxin
Endotoxin results were available for 243 reports. Of those, 30% had none detected, 55% were below the 0.5 EU/mL limit of quantification, and 15% had measurable levels between 0.5 and 40 EU/mL. The authors found that purity did not predict endotoxin (R-squared below 0.01, P=0.75) (source 1). In plain terms, a high HPLC purity did not tell the authors anything about endotoxin. That matches the point in our article on sterility and endotoxin: these are separate tests. With endotoxin data on under 4% of reports, the authors themselves say this "substantially limits" conclusions about how common it is.
Vendor spread
The 20 largest vendors supplied 3,098 samples, about 49.3% of the analyzed total. Performance varied widely between vendors, and the authors conclude that who made the product matters as much as which compound it is (source 1).
What the data cannot show
- Not peer reviewed. It is version 1 of a preprint, and the page says so.
- Not a random sample. Submissions to the testing company are voluntary, which the authors say introduces selection bias. Failure rates here do not describe the whole market.
- Thresholds are the authors' choice. Neither model is a regulatory standard.
- Narrow measurements. Only purity, abundance and endotoxin were assessed. Sterility, stability and structural integrity were not, so the authors describe the failure rates as a lower bound.
- Commercial source. The data come from a company that sells testing, and we have not independently checked them.
What a careful reader takes from it
Three habits follow from the numbers, none of which depends on trusting any one lab. First, ask which question each line on a report answers: identity, purity or amount. Second, notice when a report gives purity but no measure of abundance against the label. Third, treat any single pass-or-fail rate as a product of the cutoff chosen. The same 6,285 reports produce a 41.6% failure rate under one standard and 71.1% under the other.
To try the cutoffs yourself, the COA Bench tool includes a "Would this report pass?" check that applies both models to numbers you enter. For the document-reading basics, see how to read a COA and our line-by-line walkthrough.
Limits of this article
We read the preprint's web page, not a peer-reviewed version, and the last part of the page was not read. Numbers are quoted as the authors report them. The paper contains cost comparisons that we left out because they are not about analytical quality.
Sources
- Mendias CL, Awan TM, "Evaluation of Research Grade Peptides Marketed Directly to Consumers Reveals Extensive Variability in Purity and Measured Abundance," Preprints.org, posted April 24, 2026 (version 1, not peer reviewed), doi:10.20944/preprints202604.1748.v1: preprints.org