The peptide grey market collapsed in 9 months. Quality got dramatically better at the same time.

Who this is for: if you’re new to research peptides, a short glossary lives at the bottom of this piece. If you’ve been buying from this market for years, the vocabulary here will read native. We’ve written for both.

The headline

Between June 2025 and March 2026, the three largest vendors in the US research-peptide grey market stopped shipping.

• Amino Asylum, physically raided by the FDA at a Memphis warehouse on or around June 18, 2025. ~400,000 monthly visitors lost overnight.¹
• Science.bio, voluntarily wound down on January 27, 2026. Community framing was that the “ethical vendor” had concluded the risk/reward had inverted.²
• Peptide Sciences, voluntarily shut on March 6, 2026, with no refund process for pending orders. The closure came 8 days after Secretary Kennedy announced on JRE #2461 that the FDA was moving to restore Category 1 status for compounded peptides³, which would let compounding pharmacies legally dispense BPC-157 and TB-500, competing the “research use only” channel out of existence on the legal side.

In that same window, US Customs and Border Protection seized roughly 5,000 individual peptide shipments entering through Cincinnati alone.⁴

The market peptide buyers are operating in today is not the market of early 2025. It is post-collapse. That matters for every sentence that follows.

What we measured

TitrateLab maintains a database of every verifiable third-party peptide assay we can get our hands on. The sources feeding it include Finnrick’s 6,813-sample panel testing program,⁵ Janoshik’s public-verification portal⁶ at public.janoshik.com (~1,600 OCR’d records after a recent corpus expansion crawl of MESO-Rx’s analytical-lab subforum), MESO-Rx community uploads, and Discord drops where a buyer paid out of pocket to get a vial through HPLC.

As of this recount (July 31, 2026):

Peak-month testing volume is up roughly 10× year over year. Third-party assay has gone from a premium signal to a bare-minimum-to-ship signal in that window.

The corpus has also changed shape, not just size. When this piece was first published it was two labs, and Finnrick was most of it. Freedom Diagnostics is now 38,508 published rows, about 62% of everything, which matters for every aggregate below because the three sources do not agree with each other about dose. That disagreement is Finding three.

Finding one: quality has gotten dramatically better

If you asked an MESO-Rx regular “is the peptide market getting worse or better,” the default answer through most of 2025 was “worse.” Exit scams were in the feed weekly. Photoshopped Certificates of Analysis⁷ were circulating. The Ambrus et al. paper, published in JMIR in 2024, had tested seven randomly-purchased samples of research-grade semaglutide and found measured purity in the 7.7%–14.37% range versus the 99% claimed, with endotoxin detected in every single sample.⁸ The Ambrus finding is still the most-cited peer-reviewed primary source in the space.

We looked at the monthly rate of batches testing below 95% purity, the threshold most experienced buyers treat as the floor for consumable material:

The fail rate falls from a peak of about 6.6% in February 2025 to under 1% by mid-2026, on monthly volumes that grew by more than an order of magnitude. That is the dominant story in the data and nobody is writing it.

There is a causal explanation and it is exactly the one that the post-collapse framing implies. The weakest operators left the market, voluntarily or otherwise, during the same window that the fail rate collapsed. The regulatory wave (FDA Category 2 reclassification in September 2023⁹, the Dec 2024 warning-letter wave, the Amino Asylum raid, the Science.bio closure, the Peptide Sciences shutdown, the April 2025 ITC General Exclusion Order giving CBP seizure authority on tirzepatide¹⁰) was not just a shutdown wave. It was a filter.

The Ambrus 2024 finding remains real, and every new GLP-1 buyer should read it before their first order. But the population Ambrus sampled (randomly-purchased semaglutide from the open grey market in 2023) is not the population that is shipping in 2026. The bad operators that produced those 7.7%-pure samples have either closed, been seized, or been priced out of the market by buyers who now demand a valid Janoshik verification URL before they purchase.

Finding two: purity is rarely the problem. Dose accuracy is.

Across 50,617 certificates where we have a measured purity:

Nearly three of every four certificates come back with ≥99.5% purity. Only 1.7% are below 95%.

But there are two different measurements here. Purity is the fraction of the peptide-plus-impurities mixture that is the target compound. Dose accuracy is whether the vial actually contains as much peptide as the label claims. A vial can be 99.8% pure and still contain 30% less peptide than its label says, if the manufacturer weighed the acetate salt and called it “5 mg peptide.”

Here are the same certificates grouped by peptide (minimum 50 samples, measured against the printed label, ordered by average deviation; deviations with |value| > 50% excluded as OCR/label anomalies):

So the sign is gone, but the ordering is not, and the ordering is the part worth keeping. Melanotan II, CJC-1295 and Ipamorelin still sit at the light end of the ranking, and the last two columns show why that still matters: about one certificate in five for those compounds comes in more than 5% under label, roughly double the under-label rate of the GLP-1s. They are not underfilled on average. They are inconsistent, missing the label hard in both directions, which is a different problem with a different remedy.

The corpus-wide picture is that essentially nothing ships light on average. Of the 48 compounds with 50 or more dose measurements, 45 average over the printed label; the three that do not (VIP at −2.7%, oxymetholone at −2.8%, glutathione at −0.4%) are not among the compounds this article previously flagged. BPC-157 averages 11.2% over. Tirzepatide and cagrilintide test heavier than claimed. There are two honest explanations for the overfill, and we have not yet earned the right to pick between them:

1. Manufacturers are hedging. Add extra peptide to avoid short-ship complaints, which are easier to verify and louder than “my vial was labeled 5mg but it was actually 5.4mg.”
2. The mass measurement is catching things other than the peptide. HPLC tells you purity. It does not tell you mass. A batch that reads “5.4 mg with 99.0% purity” may have 5.4 mg of stuff, of which 5.35 mg is peptide and 50 μg is residual TFA.

Each of these happens on some batches. Which one dominates in aggregate, we don’t know yet. When the data answers that question, we’ll write it up.

Practical takeaway for Segment A readers: the “my vendor shorted me” narrative is genuinely less common than the forum surface area suggests, and the opposite miss is now the common one. Assume nothing about direction from the compound alone. The three peptides with the widest spread against label are Melanotan II, CJC-1295 and Ipamorelin, which means a batch-specific COA is worth more on those than on a GLP-1, but read it to find out which way your batch missed. Do not apply a standing upward correction to any compound in this table on the strength of its average, and do not apply one to those three at all. An earlier version of this article told you to, which was wrong.

Practical takeaway for Segment B readers: when you look at a COA, the number you care about is not only the purity percentage. Also look for “quantity,” “net peptide content,” or “mass recovery”, whatever the lab calls its measurement of actual peptide mass versus label. A 99% pure vial that contains 3mg instead of the labeled 5mg is a worse product than a 95% pure vial that contains exactly 5mg.

Finding three: the labs disagree with each other on quantity, and it matters

Three testing ecosystems now dominate our data, where two did when this piece was written. Finnrick runs panel testing against a rotating pool of aggregator labs and publishes summary reports; Janoshik is the Czech independent lab most experienced community buyers treat as the gold standard for direct verification; Freedom Diagnostics is the source that has grown fastest and is now the majority of the corpus by volume.

Finnrick reads roughly 5 percentage points lighter than Janoshik and 8 lighter than Freedom Diagnostics, on overlapping compound populations. That is not noise. The community has argued three explanations for the gap:

1. Different analytical method. Janoshik appears to include measurements that detect mass Finnrick’s aggregator labs don’t, counter-ions, residual solvents, water of hydration. Under this reading Janoshik is more rigorous and the “over-labeling” they report is real.
2. Self-selection in the submission pipeline. Vendors who believe their batch is heavy submit to Janoshik preferentially because they trust the result. This would explain higher measured deviation without implying any methodology difference.
3. Finnrick’s aggregator labs may systematically undercount over-labeling. The uncharitable reading: vendors contracting into Finnrick’s panels want to see purity, not mass, and the selected lab methods accommodate that. This would damage the credibility of aggregated datasets, including ours, that lean Finnrick-heavy. We list it as one of three competing explanations we can’t rule out, not as an assertion.

We have not resolved this. The same peptides show the same direction of gap, and the third source sits at the heavy end with Janoshik rather than splitting the difference:

Compound by compound, Janoshik reads 5 to 7 percentage points heavier than Finnrick and Freedom Diagnostics reads heavier still. A disagreement that consistent across distinct peptides and three independent sources is hard to attribute to chance, and it means the corpus-wide overfill figure in Finding two is partly a statement about which labs are supplying our data.

We have stopped publishing a single aggregate trust score based on these sources until we finish that work. The test_score column in our database is currently bimodal, it spikes at 8.0 and 10.0 rather than distributing smoothly, precisely because we weight Janoshik and Finnrick equally when they measure different things. That’s a methodology bug on our side and we’re fixing it before we publish any vendor leaderboard.

Our near-term editorial rule: when you see a quantity-deviation number in a TitrateLab post, we will disclose whether it came from Finnrick-aggregator labs or Janoshik direct verification.

What actually got tested

Top ten single-compound peptides by test volume, in descending order:

A further 8,730 certificates cover blends rather than single compounds and are excluded from this table.

Retatrutide was tested 7 times in our first month of data. In June 2026 it was tested 1,633 times in a single month. The growth is not a lab’s marketing budget, it is the post-collapse market consolidating around compounds that actually have something clinical underneath them. Retatrutide and tirzepatide are Eli Lilly pipeline molecules; cagrilintide and semaglutide are Novo Nordisk. All four carry real phase-III trial data. They are also, overwhelmingly, what the new GLP-1 segment is buying.

Finding four: there is a peptide being bought blind

Human growth hormone (“HGH,” “generic HGH,” “rec HGH”) appeared in 6,172 peptide-tagged Discord messages at the April 2026 snapshot of our message corpus. We now have 394 third-party assays on HGH across the entire span, counting every spelling the certificates use (HGH, rHGH, somatropin, human growth hormone) and excluding the 176-191 fragment, which is a different molecule. Earlier versions of this post gave 70 in the body and 96 in the key findings, which disagreed with each other; 394 supersedes both.

That’s a roughly 16× gap between how much HGH is discussed and how much HGH is tested, still the largest gap of any compound we track. It has narrowed a great deal from the 88× this article originally claimed, and most of that narrowing is testing catching up rather than discussion falling off. People are buying it on vendor-claim, with minimal verification, at scale. And the sentiment profile backs this up quantitatively: in the Tier-2-enriched subset of HGH-tagged Discord messages (n=267 peptide-relevant HGH messages processed through Claude Haiku 4.5 classification), questions outnumber positive + negative sentiment combined — 70 questions vs 34 positive vs 27 negative. Buyers aren’t confidently describing results. They’re asking whether it’s real.¹¹

Our next piece after this one is a deep dive on why HGH is the dark corner of the post-collapse market and what the 394 third-party assays we now have show.

Methodology and what we’re not saying

This dataset has real gaps. If you are using these numbers to make a purchase decision, you should know them:

• Self-selected testing. Every batch in our corpus was submitted to a lab by someone (manufacturer, vendor, aggregator, or customer) who chose to pay for a test. Batches that never made it to a lab are systematically more likely to be the bad ones. The Ambrus 2024 paper sampled from open purchase; we cannot. Our distribution is a best-of-what-got-tested, not a representative sample of what ships.
• Lab variance is real and large, as the Janoshik vs. Finnrick spread shows. The aggregate numbers in this post hide methodology disagreement.
• Temporal bias in the COA corpus. 61% of our lab data is from Q4 2025 and Q1 2026. Manufacturer behavior in early 2024 was different. Drawing conclusions about “2024 peptide quality” from this corpus is inappropriate.
• Temporal bias in the Discord + forum corpus. Most of the messages and posts in our community-intelligence pipeline come from the last few months — the Oedipus fleet’s continuous-listener infrastructure reached its current scale in late 2025 and has been real-time ingesting since. Earlier periods are covered via select targeted backfills (specific high-value forum threads, vendor-review archives, the MESO-Rx analytical-lab subforum crawl) rather than comprehensive historical ingest. Longitudinal claims about community sentiment or discussion volume that span the Q3/Q4 2025 boundary should be read as directional, not statistically complete.
• Manufacturer mapping is imperfect. Our 4,582 distinct “manufacturer” strings resolve to roughly 3,860 canonical entities, which is still far above the true underlying OEM count once the Western storefront → OEM graph is properly resolved. We are rebuilding that mapping and will publish it when it’s complete. Treat the manufacturer count as a count of storefront identities, not factories.
• No vendor-level claims in this piece. We have the data to name which Western storefronts ship the highest and lowest fail rates. We’re not naming them yet. Our editorial rule is that we do not name a vendor as worse-than-average without triangulated documentary evidence and a clean, bimodality-free scoring rubric. We don’t have the second thing yet.
• We inspected the glp1forum attachment backlog directly and it is not a hidden COA corpus. Of 229 randomly-sampled attachments from the “review” and “warning” post categories, zero were actual third-party lab reports, they were overwhelmingly Telegram/Discord screenshots, customer unboxing photos, vendor pricelists, DHL shipping pages, and fraud-impersonation warnings. Real COAs on this forum aren’t uploaded as images; they’re referenced via URLs to public.janoshik.com which our Janoshik scraper already ingests. The corpus count above is not understated by a hidden 200-600 COAs in this backlog, as we initially hypothesized.
• Janoshik public-portal images are ephemeral. The Janoshik verification platform purges PNG images from older test records unpredictably; the test IDs and structured metadata persist but the underlying image file eventually 404s. When we crawled historical data, roughly 71% of the PNGs older than a few weeks had already been purged, which means anything we didn’t OCR or download at first touch is functionally lost. We now cache every PNG locally at ingest time. All Janoshik-derived figures in this post are from records where the image was successfully OCR’d into structured fields before the purge, older records with populated data but no recoverable image remain trustworthy, but we cannot re-verify them.

We also won’t publish per-user Discord data in any article. The community sentiment numbers in this post are aggregated across the campaign-tagged corpus with no individual identification.

What’s next

In order:

• Methodology page, the rubric, the thresholds, the Janoshik weighting, open for community critique before we grade anyone. This is a prerequisite for any vendor-named content.
• The HGH investigation: 6,172 mentions, 394 COAs. A piece that names no vendors as scammers but names the safety gap directly. [Now published, see HGH is discussed far more than it’s tested.]
• “Why Janoshik’s numbers disagree”, the lab-variance deep-dive, with sampled batch-level comparisons.
• MAHA / PCAC July 23 preview, time-boxed explainer on the regulatory reversal. What happens if BPC-157 and TB-500 get moved back to Category 1. Written before the July 23 review, not after. [Now published, see The July 23 PCAC review.]
• The Peptide Vendor Graveyard, a permanent, dated, evidence-backed archive of every verified vendor closure, FDA action, and incident 2018 → present. Canonical reference. Backlinks of record. [Now published, see The Peptide Vendor Graveyard.]

Data current as of 2026-07-31. This post will be revised as new batches land and as our methodology evolves. Corrections welcome through the TitrateLab community link in the footer. Every query behind these numbers is reproducible from the methodology page when it publishes.

Corrections and updates

• 2026-07-31 — Correction, and a withdrawn recommendation. This article reported that Melanotan II (−4.5%), CJC-1295 (−2.5%) and Ipamorelin (−2.0%) “consistently ship light,” and advised readers stacking them to “re-dose by the COA, not the label.” That advice is withdrawn. Two separate errors produced it. The first is ours: the deviation figure was taken from a column that compares the assay to the vendor’s batch claim where one is published, and a batch claim is typically an advertised overfill, so a vial running over its printed label could book a large negative. Recomputed against the printed label, the same Finnrick certificates that gave −4.5% for Melanotan II give −1.3%. The second is corpus growth: at 62,079 published certificates, with two large sources that read heavier than Finnrick, all three compounds now average over label (+6.8%, +4.6%, +6.3%). Advising an upward dose correction on an already-overfilled product is the most harmful thing this article could have said, and it said it for three months. The compound ranking survives the recount and the sign does not; the section has been rewritten around dispersion, which is what the data actually supports.
• 2026-07-31 — Correction, the 8× fail-rate claim. The article claimed an “8× improvement in the sub-95% fail rate over 14 months.” It no longer reproduces. The multiple was anchored on a January 2025 containing 55 tests at 10.2% sub-95%; backfill has taken that month to 350 tests at 2.3%, so the starting point was an artifact of thin early coverage rather than a property of the market. The same live data supports multiples between roughly 1.4× and 9× depending on the anchor month, which is why the multiple has been dropped in favour of the levels. The improvement itself is real: monthly sub-95% peaked near 6.6% in February 2025 and runs under 1% through mid-2026.
• 2026-07-31 — Data, corpus recount. Every aggregate in this piece was frozen at the 8,452-batch snapshot of April 2026, and an editor’s note dated April 25 updated that to 11,227. Both are superseded: the corpus is 62,079 published certificates across roughly 3,860 resolved manufacturers. The purity distribution, the monthly fail-rate table, the per-compound dose table, the lab comparison and the test-volume ranking are all recomputed. The social card, which read 8,452, is corrected. The URL slug still reads 7007, the published-certificate count on the original publication date; we have not changed it, because changing a published slug breaks every inbound link and citation to this piece for a cosmetic gain.
• 2026-07-31 — Correction, HGH assay count. The key findings said 96 third-party HGH assays, the body and the closing roadmap said 70, and an 88× discussion-to-testing ratio was printed alongside both, which was not consistent with either. Counting every spelling in the certificates and excluding the 176-191 fragment, the live figure is 394, and the ratio against the 6,172-message snapshot is about 16×. The gap is still the largest we track.
• 2026-07-31 — Correction, lab sources. This article described the corpus as two primary lab providers. Freedom Diagnostics is now 38,508 published rows, about 62% of everything, and it reports dose deviations heavier than Janoshik. Presenting a Finnrick-versus-Janoshik comparison without naming the largest source understated how much the corpus-wide dose figures depend on source mix. Finding three now compares all three.

A short glossary

• COA / Certificate of Analysis, the lab report a vendor publishes claiming their batch is pure and correctly dosed. Forgeable if unverified; reliable when linked to a lab’s own verification portal.
• HPLC, high-performance liquid chromatography, the standard method for measuring peptide purity.
• Purity vs. quantity deviation, two different measurements. See above.
• RUO / Research Use Only, a label on grey-market peptides that is not a legal defense. The FDA has explicitly called it “a ruse” in warning letters.¹⁴
• Janoshik, a Czech lab (not ISO/IEC 17025 accredited) that most of the experienced community treats as the gold standard anyway. Their public.janoshik.com verification portal is the hardest thing in this market to forge.
• Finnrick, an independent peptide testing program that publishes a letter-grade vendor ranking based on 6,813 samples across 204 vendors. Data-adjacent competitor to TitrateLab with a different methodology and a narrower scope.
• MESO-Rx, the oldest surviving forum for bodybuilding + peptide discussion (thinksteroids.com/community). Slow, moderated, memory-rich. If a vendor has a 5-year-old megathread there, that is the closest thing to a vendor credit score this market has.
• Segment A / Segment B, our internal shorthand. Segment A: experienced buyers who read chromatograms. Segment B: the GLP-1 / weight-loss wave who mostly can’t.

TitrateLab is the intelligence layer for the peptide and grey-market underground. COAs, pricelists, exit-scam early warnings, vendor grades, clinical research, updated live. Join the waitlist to be notified when the platform opens.