George Akerlof published "The Market for Lemons" in 1970 in the Quarterly Journal of Economics. It runs to thirteen pages and won the Nobel Prize in 2001. The model is simple enough to fit in a paragraph, and the peptide market is now running through it in front of us.
Akerlof's setup: a market for used cars. Sellers know whether a given car is a "peach" or a "lemon." Buyers cannot tell the two apart before purchase. Buyers therefore pay a price that reflects the average. Sellers of peaches will not accept that average, because their car is worth more. They withdraw. The average quality on offer drops. The price drops to match. The next tier of honest sellers withdraws. The cycle continues until the market either clears at the bottom or fails to clear at all. The technical name is adverse selection. The shorter version: when verification fails, the bad pushes out the good.
The peptide market in 2025 and 2026 is the textbook version, with one complication Akerlof did not have to model: the lemon can put the buyer in the hospital.
How the model maps onto a peptide vial
A buyer wants a 5 mg vial of a research-grade peptide. They cannot weigh it. They cannot run mass spectrometry on it. They cannot tell, looking at the seal, whether it was filled in a clean room or a garage. They are buying on signals: a website that looks professional, a published certificate of analysis, a Reddit thread that says the vendor has shipped reliably for eighteen months.
Each of those signals can be manufactured for less than the margin on a single shipment. A clean website is a Shopify template. A certificate of analysis can be lifted from a different vendor's site, rebranded, and posted. The Reddit thread can be bought. The labor cost of building the surface area of a credible peptide vendor is, conservatively, four figures. The revenue on the first month of a Reddit-promoted launch is, conservatively, six figures. The asymmetry in Akerlof's paper was about used cars worth a few thousand dollars apiece. The asymmetry here is much larger.
Run his model on those numbers and the prediction is unambiguous: the price clears at the worst plausible quality the buyer is willing to risk. Honest vendors who paid for ISO 17025 lab partnerships, real chain-of-custody documentation, and slow-burn community reputation cannot undercut the substituted-vial price. They either exit, raise their price into a niche the broader market does not reach, or reduce their own quality to stay competitive. Each of those moves makes the average worse.
Why more reviews do not fix it
The instinctive response to a lemons market is to add more signals. More reviews, more ratings, more side-by-side comparisons. The problem is that reviews are themselves a market with the same asymmetric information.
There is a documented going rate for a Trustpilot review with a verified-purchase badge. There is a labor market for aged Reddit accounts. There is a brokerage tier above both of those where coordinated review rings get bought and sold across subreddits. None of this is hypothetical. It has been written about in the trade press for a decade. The signal-to-noise ratio of consumer reviews is low enough that the rational buyer treats them as advertising, which is what they have, in fact, become.
What this means is that the buyer's information set is not improved by adding more reviews. It is improved by adding a class of signal that the seller cannot pay to alter. That is the entire point of third-party attestation, and it is the only structural answer Akerlof's model permits.
What "third party" actually has to mean
Not every third party will do. A verifying institution that takes seller money for placement, accreditation, or removal is not a third party in the model. It is a four-way trade with extra steps, and the buyer's information set is no better than before.
The institutional shape that survives the model is narrower than most readers expect. Four conditions matter, and a system that fails any of them collapses back into the original problem.
The lookup has to be free at the point of use. Paywalled verification is gatekeeping with a marketing budget attached. A buyer who has to subscribe to find out whether a vendor is verified will not subscribe, and the verifying signal does not reach the price.
The lookup cannot require an account. Identity-gated verification creates a chilling effect that suppresses the very evidence the registry depends on. If buyers have to log in to see the record, the operator of the registry gains a commercial dataset of the people checking it, and that dataset becomes itself a valuable asset that can be sold or compromised.
The record has to be append-only. A verified party who can pay to delete their incident history can pay to delete it the moment it matters most. A registry that allows deletion is a review site with a courthouse aesthetic.
Verifications have to expire. ISO 17025 accreditations cycle on a two-year clock. Domain ownership changes hands. A "verified" mark with no expiration date is silently lying as soon as the underlying conditions move, and most do, within twelve months.
A registry that meets all four conditions can produce a signal that is structurally hard to fake, because faking it requires either compromising the registry itself (a binary, attackable event) or producing the underlying documentation, which is what the buyer wanted in the first place.
The Carfax analogy is the point
The used-car market did not solve its lemons problem by adding more reviews. It solved it by inventing a vehicle history report that was free to view, tied to a single permanent identifier (the VIN), and operated by a party with no stake in any individual sale. The price of a used car with a clean Carfax now reliably exceeds the price of one without, by an amount that closely tracks the cost of obtaining the report. Akerlof's model, run in reverse, recovers the price spread.
The peptide market needs the same thing, applied to lots and vials rather than chassis numbers. The infrastructure is older than the internet. The only question is which institution builds it under conditions that do not rebuild the asymmetry the moment they are codified.
If a reader takes one thing from the 1970 paper, it is this: the cure for an information failure is not more information. It is the right institutional shape of the information that already exists.