Blind testing and why it matters for peptides

What does blind testing actually mean?

Lesson 11 showed that one vial is a sample of its batch. The next question is who picked that sample, and who knew where it came from. "Blind" means someone in the chain is deliberately kept from information that could change what they do. In peptide testing that applies to two people.

• The seller. If the seller knows a vial will be tested, the seller can decide which vial it is. Seller-blind sampling means buying the vial the way any customer would, so the seller does not know it is a test.
• The analyst. If the analyst knows whose vial it is, expectations about that seller can creep into judgement calls. Analyst-blind testing means the lab receives a coded sample with no seller name.

The word also gets used a third way. Some services offer a "blind" identity screen, where the lab is not told which of several molecules to expect and must work it out from the data. That is useful, and relies on the mass spectrometry in lesson 7, but it is a different thing. When someone says a vial was "blind tested", ask which of the three they mean.

Why does it matter who chooses the vial?

A seller who knows a vial is going to a lab has every reason to send the best one. The report can then be genuine, verifiable and accurate and still describe a vial unlike what customers receive.

The World Health Organization's guidelines on surveys of medicine quality compare overt sampling, where the seller knows, with a covert mystery-shopper approach. They note that "if the seller knows or is concerned that his or her stock contains illegal or poor-quality medicines and that the buyer is potentially linked to the NMRA, this may influence which medicines are offered", and that in such cases "the identity and purpose of the buyer should not be generally known by the outlet being evaluated" (WHO TRS 996, Annex 7). The mystery shopper "mimics a normal shopper", records the purchase details afterwards, and each sample gets its own code.

Researchers studying online sellers use the same idea. A 2024 study of semaglutide sold online without a prescription combined market surveillance with "targeted product test purchases", then ran visual inspection, sterility, endotoxin and LC-MS analysis on what arrived (Ashraf et al., J Med Internet Res, 2024). The vials analysed were the ones the sellers actually shipped to a buyer.

Regulated manufacturing has its own name for the opposite habit. The US FDA describes "testing different samples until the desired passing result is obtained" as "testing into compliance" and calls it inconsistent with good manufacturing practice (FDA, Data Integrity and Compliance With Drug CGMP). A seller who sends several vials to a lab and publishes only the best report is doing something similar, whether or not they think of it that way.

Why should the analyst not know the seller?

Most of an analysis is mechanical, but not all. Someone decides where a peak starts and ends (lesson 6) and whether a mass is close enough (lesson 7); lesson 13 covers more of these judgement calls. Each is an opening for expectation to shape the answer.

Forensic science has studied this closely. A paper on bias mitigation in forensic casework defines cognitive bias as decision patterns that occur when people's "preexisting beliefs, expectations, motives, and the situational context may influence their collection, perception, or interpretation of information". It describes how Costa Rica's Department of Forensic Sciences piloted blind verifications, case managers and less exposure of examiners to task-irrelevant information (Alfaro et al., Forensic Science International: Synergy, 2025). The point is not that analysts are dishonest. Knowing who is affected by a result is irrelevant to the chemistry, and removing it removes one way to be wrong.

The effect has been measured. In a 2020 forensic toxicology experiment, participants (undergraduate students, a limitation the authors acknowledge, in small groups) read the same immunoassay data with or without case background. With no context, error rates on the clear-cut cases were 6% to 12.5%; with context, even context that pointed towards the correct answer, they were 15% to 19% (Hamnett and Dror, Forensic Science International: Synergy, 2020). Earlier, when 17 expert DNA examiners interpreted data from a real case without its background, most disagreed with the original laboratory's conclusion, which the authors read as a sign that the case context had influenced that conclusion (Dror and Hampikian, Science and Justice, 2011).

There is a commercial side too. A lab that knows a large seller sends it regular work has an interest in keeping that seller happy. It cannot favour a client it cannot identify.

What can and cannot be hidden?

The lab must be told the compound and the label amount, or it cannot check identity against the right mass or amount against the right label. Those are facts about the vial, not the seller. What is hidden is anything that could identify the source.

You can often tell from the paperwork which kind of sample a report came from. WHO survey guidance has the certificate of analysis carry both the laboratory's own sample reference number and the sample code assigned at collection, and has buyers and sellers entered in the survey database "using codes only" (WHO TRS 996, Annex 7). A report whose client or sample field shows a seller's brand was labelled, because the analyst could see it. A coded report shows a sample code distinct from the seller's batch code, with a separate published record that links the two.

Seller blinding can also fail. A small seller who sees the same delivery address or name on repeated orders may guess. That is why a careful testing programme varies its buyers and delivery addresses, and why one anonymous purchase is never treated as conclusive on its own.

How does this change how much weight I give a report?

From most to least protected against selection:

1. Anonymous purchase, coded sample. The seller did not know, and the lab did not know whose vial it was.
2. Buyer-submitted, coded sample. The seller did not choose the vial, but the buyer knows the source. Good evidence about that vial.
3. Seller-submitted, coded sample. The lab was blind, but the seller chose the vial. The result is genuine for that vial and says less about what ships.
4. Seller-submitted, uncoded sample. The seller chose the vial and the lab knew who sent it.

None of these makes a report false; they change what it can stand for. A seller's own certificate, even a verified one, is level 3 or 4 at best, because the seller chose what to send. The peptide seller red flags guide covers the other questions to ask about a seller's documents, and the legit checklist asks whether a seller has ever had a third-party test published on a vial it did not choose.

What happens after the result comes back?

The signed report comes back carrying only the sample code. Tested.ph then matches the code to the seller and the order, and the record is published with both. The stored report also carries a SHA-256 fingerprint, so a changed file would no longer match the record. The full chain of custody is on how it works.

For tirzepatide, identity is checked by LC-MS, so the lab's only question is whether the mass matches tirzepatide. It has no way to know whose vial it is.

Check your understanding