Peptide blind testing: how to verify a lab is not fabricating results

• Blind testing means sending an unlabeled sample to a lab and asking it to characterize what is in the vial without being told the expected answer.
• A split-sample variant — one aliquot to lab A with the label, another to lab B blinded — is the strongest form of inter-lab verification a buyer can run.
• A pass means the blinded lab named the molecule from LC-MS, reported purity and content from raw chromatograms, and matched the labeled identity within method tolerances.
• Common ways labs fail: refusing unknowns, returning a percentage with no underlying data, mass-spec results that don't resolve isobaric analogs, or purity figures that drift between submissions of the same sample.
• If a lab will not accept blind submissions, that itself is the answer. A characterization pipeline that cannot identify unknowns is not a characterization pipeline.

Every peptide on the research market ships with a COA. Most of them say the same things: 99% purity by HPLC, identity confirmed by LC-MS, sometimes a content figure, sometimes an endotoxin number. The format is so standardized that a competent designer can produce a convincing-looking COA in an afternoon — and that is exactly what happens in the grey market.

A vendor-issued COA is unverifiable by definition. The seller controls the sample, the lab relationship, the methods reported, and the numbers printed. A third-party COA is better, but only if you can verify that the third party actually tested that vial and not a different one. Even a real independent lab can be fed an aliquot of a clean reference batch while the production batch ships separately.

Blind testing removes those degrees of freedom. The buyer controls the sample. The lab does not know what answer to write. The result has to come from the data, because there is no other place for it to come from.

Not every blind test is equally strong. The three variants below trade convenience for evidential weight. Pick the strongest one your budget supports.

Protocol 1: Single-blind submission

Decant or aliquot a sample from a vial you bought, place it in an unmarked container, and send it to an independent lab. The lab is told only the requested assays (identity, purity, content); the molecular identity, vendor, and batch are withheld. The lab returns a characterization report on the sample as received.

This is the lightest-weight version and the easiest to run. It proves the lab can identify and quantify an unknown peptide. It does not prove that any specific vendor batch matches its label, because you have only one report to compare against.

Protocol 2: Split-sample inter-lab

Split a single vial into two aliquots. Send aliquot A to lab A under its true label. Send aliquot B to lab B blinded. Compare the two reports for identity, purity, and content. If both labs converge on the same molecule and the purity numbers agree within method tolerance (typically ±1–2 percentage points for RP-HPLC at >95% purity), the labs are corroborating each other. Divergence is informative regardless of direction.

This is the protocol regulated pharma uses to qualify a new contract lab, and it is the strongest verification a non-regulated buyer can practically run.

Protocol 3: Spiked or substituted blind

A more aggressive audit: instead of a real production aliquot, send the lab a deliberately prepared sample. Options include (a) a known reference standard with a known purity, (b) a binary mixture of two related peptides at a known ratio, or (c) a blank vehicle with no peptide at all. A competent lab returns answers consistent with what you put in. A marketing lab returns whatever the COA template defaults to.

This protocol is the closest thing to a controlled experiment a buyer can run on a lab, and it is what we recommend for anyone considering a long-term testing relationship with a new analytical partner.

A blind characterization that holds up under scrutiny has six features. Any of them missing is a red flag; all six absent is a fabrication.

• Identity from raw data, not from the request. The report names the molecule based on the LC-MS spectrum — observed monoisotopic mass, charge-state envelope, isotope pattern — not based on what the buyer asked it to confirm.
• Purity from an attached chromatogram. A purity percentage with no chromatogram is unfalsifiable. The chromatogram should show the integration baseline, retention times, and named impurity peaks where present.
• Content figure from a quantitative method. Amino-acid analysis remains the reference method for peptide content. UV-based quantitation against a calibrated standard is acceptable for residues with strong chromophores. Either way, the method has to appear in the report.
• Method transparency. Instrument model, column chemistry, mobile-phase composition, gradient program, flow rate, injection volume, detector wavelength, ionization mode. Enough that another competent lab could reproduce the run from the report alone.
• Tolerances stated explicitly. Mass accuracy in ppm, purity integration threshold, content recovery range. A report that gives results without uncertainty is not analytical chemistry.
• Sample chain of custody. When the lab received the sample, in what condition, how it was stored, what the test article was labeled as on arrival. Closes the gap between “a sample was tested” and “this sample was tested.”

The failure patterns are consistent enough to use as a diagnostic. If you run a blind test and see one of these, the lab is not performing the work it advertises.

• Refusing the submission. “We need the molecular identity to test it.” Identity is what LC-MS is for. A lab that cannot start without the answer is not running LC-MS.
• Returning the buyer's expected answer. On a substituted-blind (Protocol 3), the lab reports the molecule the buyer told the vendor it was, not what was actually in the vial. This is the strongest possible evidence that the report is templated, not measured.
• Inability to distinguish close analogs. Peptide pairs that differ by a single methyl or a single amino acid — semaglutide vs. retatrutide fragments, BPC-157 vs. BPC-157 arginate, GHRP-2 vs. GHRP-6 — require LC-MS with adequate resolution and an analyst who reads the spectrum. Reports that confuse these are a sign of a low-resolution instrument or a non-analyst pipeline.
• Purity figures that drift. Submit the same sample twice, a month apart, blinded both times. Real analytical methods reproduce within method tolerance. Drifting numbers indicate the report is being generated rather than measured.
• Suspiciously round numbers. 99.0% purity on every report, every batch, every peptide is a template default, not an integration result. Real chromatograms produce numbers like 97.3% or 99.6%.
• No raw data on request. A credible lab will release the chromatogram, the mass spectrum, and the integration table on request. A lab that cannot produce raw data does not have raw data.

Most buyers will never run a blind test themselves. The point of understanding the protocol is to know what to ask for — and to recognize the signals that distinguish a supplier whose data would survive a blind test from one whose data would not.

• The supplier publishes batch-specific reports, not a template COA reused across batches. Real testing produces different numbers on different batches. A library of identical-looking reports is a library of one report.
• Reports include the raw chromatogram and mass spectrum, not just a summary table. Raw data is the difference between a measurement and a claim.
• The testing lab is named and independent. The lab cannot be a subsidiary, a sister company, or an in-house QA group operated by the vendor.
• The COA is verifiable after issuance. A verification URL, a cryptographic anchor, or a public batch index that the buyer can independently confirm matches the report on the vial.
• The supplier's lab partner accepts blind submissions. Ask the question directly. A lab that runs blind work for other customers will say so without hesitation.

CertikLabs publishes every verification at a public URL tied to the specific vial, accepts blind and split-sample submissions as a standard intake type, and treats the chromatogram and mass spectrum as part of the report rather than supporting data held back from the buyer. The full verification model is described in why third-party peptide testing matters and how to tell if your peptide is actually third-party tested.

The practical workflow for a buyer who wants to verify a peptide they already bought:

1. Decant a representative aliquot from the vial into a clean, unlabeled container. For lyophilized material, transfer a portion of the powder under low humidity. For reconstituted solution, pipette an aliquot into a fresh vial.
2. Submit through the standard sample intake without naming the vendor or stating the expected identity. Specify the assays you want (typically identity by LC-MS, purity by HPLC, and content if quantitation matters for your use).
3. Receive a verification with a public URL. The report names the molecule, reports purity and content, and attaches the chromatograms and spectra.
4. Compare against the vendor's COA. Identity match, purity within method tolerance, and content within stated range together constitute a pass. Divergence on any axis is the information the test was designed to surface.

A split-sample variant follows the same workflow, with a second aliquot sent to a different lab under its true label. The two reports become the comparison.

Every step of the blind-testing protocol above can be used to audit CertikLabs. That is intentional. A verification layer that cannot withstand its own audit protocol is a marketing layer, and the entire point of independent testing is to give buyers something they can verify without trusting either the vendor or the lab.

If you are a vendor whose supplier relationship depends on credible third-party data, run the blind test on your current lab before you need to. If you are a buyer who has been relying on vendor COAs, run it on the next batch. Either way the information you get back is the information that should have been there all along.