The Molecule Is Clean. The Dose Is a Coin Flip.

A 22,271-certificate audit of the injectable grey market finds a trade mid-pivot — from anabolic steroids to weight-loss peptides — getting chemically purer even as it grows less reliable about the dose in the vial.

Drawn from the largest public archive of grey-market certificates of analysis. Snapshot: 22,271 certificates, June 2026. The methods and their limits are at the end — read them. This is partly reconstructed data, and what it can and cannot say is half the story.

This report draws on the largest public archive of grey-market drug testing ever assembled: 22,271 third-party certificates of analysis, from 2017 to 2026, covering the peptides, weight-loss drugs, and steroids that people buy outside the pharmacy and send to independent labs to verify. Read end to end, it captures a market in the middle of a transformation — and carrying risks most of its own buyers never see.

The one-line takeaway: the injectable grey market is pivoting from anabolic steroids to GLP-1 weight-loss drugs, and it is getting chemically purer even as it grows less reliable about the labeled dose in the vial. Cleaner molecules; less trustworthy doses.

What the data shows:

• A regime change. GLP-1 weight-loss drugs went from essentially 0% of testing in 2021 to 53% in 2026, while anabolic steroids fell from roughly 80% to under 5%. One in two grey-market lab tests today is a weight-loss drug.
• Purer, but worse-dosed. Average purity climbed toward 99.5% — yet the share of vials testing at least 10% under their label nearly doubled (7.5% → 13.2%). A "99% pure" badge cannot detect a half-empty vial, and 79 certificates were at least half-empty at high purity.
• Almost no one checks whether it is sterile. 97% of certificates carry no endotoxin (sterility) result at all; where it was tested, roughly 1 in 12 failed — and 11 of the 12 NAD+ batches anyone tested came back contaminated.
• A market that runs years ahead and turns over in weeks. Buyers tested retatrutide roughly three years before any approval; half of all 577 compounds in the archive appear only once; the median vendor lasts 29 days.
• One lab; China-made, Western-fronted. A single overseas lab, Janoshik, underwrites 57% of the trust in this market, and the certificates almost always name the Western reseller rather than the Chinese factory behind it.

One caveat up front, because it shapes everything else: this is a record of what we could discover, not a census — roughly a third was reconstructed from archived and re-scraped data — so every trend here is a within-year share, never a raw-growth claim. The methods and their limits are at the end, and they are part of the story.

What this archive actually is

A certificate of analysis is the receipt of trust in an unregulated trade. A buyer ships a vial to a lab; the lab reports the compound, its purity, and its milligrams; the seller posts the result. No regulator asks for any of it. It happens because, absent a regulator, a lab report is the only credibility a grey-market vendor can buy.

This corpus collects 22,271 of them, with test dates running from November 2017 to June 2026. One lab carries most of the weight: Janoshik ran 57.3% of every certificate here — 12,763 of them. Most of the rest pass through a single aggregator that republishes several labs under coded names, so the market looks multi-lab even though the community-posted stream has standardized on Janoshik almost completely. A single overseas lab has become the de facto trust backbone of a global underground pharmacy.

The first surprise is the filing label itself. This is kept as a "peptide" archive, but nearly one in five certificates — 19%, or 4,219 — is not a peptide. Anabolic steroids alone are 15.2% (3,380 certificates), the second-largest drug class in the corpus, behind only the weight-loss peptides. The same testing pipeline serves bodybuilders, dieters, and longevity hobbyists without distinction. Three drugs — retatrutide, tirzepatide, and semaglutide — account for more than 40% of the archive between them.

The Great Inversion

The loudest pattern in the data is a changing of the guard.

Chart each year's certificates by what share went to GLP-1 weight-loss drugs and the line goes nearly vertical: essentially 0% through 2021, then 1.4%, 8.5%, 25.5%, 49.2%, and 53.2% in 2026. One in two grey-market lab tests today is a weight-loss drug. Anabolic steroids and hormones ran the opposite way over the same years — from roughly three-quarters to four-fifths of all testing through 2022 down to 4–5% by 2025–2026. The two lines cross in 2024. By 2025, not one anabolic steroid appears among the year's ten most-tested compounds, a category that had owned that list, unbroken, for five years.

The clientele changed: from bodybuilders verifying underground gear to dieters verifying compounded weight-loss drugs.

The jockeying inside the new regime is just as sharp. Retatrutide passed tirzepatide as the single most-tested compound in the third quarter of 2025 and has widened the gap every quarter since. And semaglutide — the molecule that opened the GLP-1 era — is fading inside the grey market even as the category booms around it, slipping from 153 certificates in a quarter to 19. The underground has already moved to the next, stronger thing.

What these numbers are: each is a within-year share — GLP-1 as a percentage of that year's certificates — never a raw count. This corpus is partly reconstructed history (see Methods), so absolute volume by year measures how much we could recover, not how much the market did. Shares survive that; raw growth does not. We do not claim "testing grew N-fold," and neither should anyone citing this. The GLP-1 surge in particular was checked against the live, contemporary stream and holds — a real shift, not an artifact of how the archive was built.

The central paradox: purer molecule, worse dose

Here is the finding that breaks the buyer's mental model. Purity and dose are independent measurements — is it the right, clean molecule? and did the vial hold the labeled milligrams? — and in this archive they are moving in opposite directions.

Purity is climbing. Among the live, contemporary certificates, average purity rose from about 98.2% to 99.4% between 2023 and 2026, and the share testing at 99% or better went from roughly half to 91%. The chemistry keeps improving.

The dose is slipping. Over the same window, the rate of vials testing at least 10% under their labeled dose roughly doubled — 7.5% in 2024 to 13.2% in 2026 — and severe shortfalls, 25% or more under, rose from 1.4% to 3.6%. This is the single cleanest trend in the corpus: dose-deviation data exists only on live, freshly-ingested certificates, so it carries none of the reconstruction caveats that shadow the volume figures.

The two measurements have come apart, and one cut of the data shows it cold: of 10,376 certificates that tested at 98% purity or higher, 1,121 — 10.8% — were still at least 10% underdosed, and 79 were at least half-empty. A 99% purity badge, the number buyers fixate on, cannot see a half-dose. It was never measuring that.

The counterintuitive twist: the hyped weight-loss drugs are the cleaner end of the market. GLP-1 agonists average 99.48% purity to the older research peptides' 98.63%, with a fifth as many sub-95% failures (0.78% versus 4.32%) and half the underdose rate (9.0% versus 18.2%). The gold-rush compounds drew high-volume sellers competing on public lab proof; the quieter, older peptides get the sloppier quality control.

Who pays the underdose tax

Underdosing is not the work of a few villains — it is a baseline. Across the vendor-attributed certificates, 11.7% (1,314 of 11,190 with a dose measurement) came in at least 10% short, and that shortfall is diffuse: the five biggest contributors together account for under a fifth of it. There is no single bad actor to indict. There is a background rate.

What concentrates the risk is the compound, not the seller. The older, quieter peptides run short far more often than the headline weight-loss drugs:

Underfilled by ≥10%, by compound: CJC-1295 26.0% · GHK-Cu 25.9% · Melanotan II 24.6% · Tesamorelin 18.1% · Ipamorelin 17.2% — against Retatrutide 9.0% and Tirzepatide 8.7%.

The sharpest case is GHK-Cu, a cosmetic-longevity peptide. It is the purest compound in the archive — 99.8% average purity, 99% of vials at 99%-plus — and simultaneously among the most short-changed, with 25.9% underfilled by 10% or more. Fifty-milligram-labeled vials assaying at 13 to 16 milligrams. The cleanest molecule, the skimpiest pour.

Where a pattern is robust — many certificates across many production batches — it is fair to name. Pharma Grade Peptides is the clearest: across nine distinct batches its vials tested 98–100% pure yet ran badly short, including a CJC-1295 labeled 10 mg that held about 1.1 — an 89% shortfall at near-perfect purity, the whole paradox compressed into one vendor. Others carry the same multi-batch signature at lower intensity: Dankang Biohealth (32% short across 17 batches), Bfflink (32% across 33), Limitless Life Nootropics (82% across 5), Pure Rawz (45% across 8). Patterns, not accidents.

The discipline cuts both ways. Several vendors post eye-popping "90% underfill" rates that dissolve into a single batch tested over and over — a lab artifact, not a practice — and they are deliberately left unnamed here. Some sellers run consistently over label, not under. And the market as a whole errs generous: the median vial tests +3.8% over its label, with roughly 29% overfilled by 10%-plus against about 12% underfilled. The typical grey-market vial gives you a little extra. It is the growing, compound-specific shortfall tail — not the median — that should worry a buyer.

A good average isn't a guarantee. The number that protects a buyer is not a vendor's mean dose — it is the spread batch to batch. Take Shanghai Sigma Audley's retatrutide, measured by a single lab across sixteen production batches: it averages a healthy +8.7% over label, yet individual vials in the run test anywhere from −40.6% to +22.4%. A buyer reading "averages a little over" would never guess that one batch held barely half the dose. The mean says reliable; the range says coin flip. (Because batch-level testing reaches only the well-certified ~46% of the market, this is a statement about the thoroughly-tested vendors — the ones we know the most about, not the worst-documented ones.)

Is the dose getting better or worse?

Both — and the answer flips depending on the molecule. There is no single quality trend, only a per-compound one, and to read it honestly you can only compare the two years with enough independent tests to judge each molecule on its own: 2025 and 2026. (Earlier years are too thin per compound, and the deeper history is mostly back-discovered — a curve about what we recovered, not what the market did.) Set side by side, the picture splits. The headline GLP-1 drugs drifted slightly the wrong way on dose even as the market grew chemically cleaner overall; meanwhile a few older peptides that had been chronic short-fillers pulled sharply back toward label. Quality here is not a tide that lifts or drops every boat at once — it is a per-seller, per-molecule fight, and the direction reverses depending on which vial you pick up.

Is it even safe to inject?

Purity and dose are the two questions buyers ask. There is a third they almost never get to ask, because the paperwork almost never answers it: is the vial sterile?

These are injectable drugs. A contaminated vial — bacterial endotoxin, the fever-spiking breakdown product of dead bacteria — is a clinical problem no purity figure can detect. And the corpus is nearly silent on it. Only 659 of 22,271 certificates — about 3% — carry a real endotoxin result at all. Ninety-seven percent of grey-market certificates never test injection sterility. It is, by a wide margin, the single most-flagged defect in the entire archive: "no endotoxin test" is noted on 10,457 certificates, more than three times the next-most-common concern.

Where sterility was tested, roughly 8% failed — counting a certificate as a failure when its result reads detected, borderline, or fail, or when a quantitative reading lands at 10 EU/vial or above. That risk is not spread evenly. It concentrates in the injectable weight-loss drugs and one longevity compound:

• Tirzepatide — 12.3% of tested vials failed (n=171). The most-injected new drug in the corpus is also the most likely to come back contaminated when anyone checks.
• Retatrutide — 3.7% (n=273).
• GHK-Cu, Semaglutide, Tesamorelin, Ipamorelin — essentially 0% across the batches that tested.

And one compound stands alone: NAD+, injected at very high milligram loads, came back contaminated on 11 of the 12 batches anyone tested — readings as high as 984 EU/vial, orders of magnitude over any plausible injectable limit. That is eleven of twelve who tested, not eleven of every NAD+ vial sold (a hundred and fifty-odd NAD+ certificates skip the sterility test entirely) — but for a compound people push into their veins by the hundreds of milligrams, even a small tested sample failing this completely is a flare worth naming.

A fair companion question: did we find vials of the wrong drug entirely — testosterone sold as a peptide, a stimulant in a weight-loss vial? We went looking. Essentially no. Beyond five vials sold as 5-to-60-milligram weight-loss doses that an independent lab measured at zero milligrams of active drug — true vials of nothing, though just five of 7,330 GLP-1 dose tests — the handful of "wrong molecule" hits all trace to ingest mis-tags — a steroid certificate whose label field got auto-filled with the wrong name — not to a buyer dosing the wrong compound. There is no defensible "you'll get a different drug" rate in this data. The documented sterility gap is real; a substitution epidemic is not.

The supply chain behind the brand

The certificate almost always names the reseller, not the factory.

About 98% of certificates carry a vendor identity, yet only 8–9% show an explicit China-origin fingerprint — a .cn or 163.com address, a Chinese city, a "Co., Ltd." The most common contacts are Proton Mail (598) and Gmail (463) handles; the Qingdao and Guangzhou factory accounts surface only now and then (shsry.com.cn on 103 certificates, qianyecao.cn on 52). The structure is well understood — overwhelmingly China-manufactured product, fronted by Western-branded resellers whose Proton address is the only identity the paperwork captures. The 8–9% is a floor on China origin, not an estimate of it. (Those handles are supply-chain signal, not private information; they are how a handle-only vendor is known to its buyers.)

The frontier and the churn

If you want to know what a drug regulator will be arguing about in three years, watch the grey market now. This archive tested retatrutide for the first time on October 15, 2022 — seven weeks before tirzepatide showed up, and roughly three years before any approval anywhere. The underground was dosing the next-generation obesity drug while the current one was still launching. The bleeding edge runs further still: a Lilly Phase-1 molecule, eloralintide, first appears here in May 2026; a senolytic peptide built to clear aging cells, FOXO4-DRI, in March 2025. (To be precise, these are first seen in our corpus — a discovery date, not a claim about when the compound first existed anywhere.)

The catalog is enormous and mostly one-offs. The archive spans 577 distinct compounds, and 277 of them — 48% — appear on exactly one certificate. That long tail is the exotica: Russian bioregulator peptides, senolytics, antiparasitics, a single curious buyer's single curious test. Half the chemical diversity of the grey market is things almost nobody is actually buying.

The sellers churn even faster than the chemistry. Track each vendor from its first certificate to its last and the median lifespan is 29 days. More than half — 52.8% — show up in a single calendar quarter and are never seen again, and only 6.4% persist for two years or more. This is a market of transient storefronts, not durable brands: most names that appear in the data are gone before a buyer could build a season of trust in them. The certificate proves the batch; it cannot promise the seller will still be there next quarter.

When buyers do combine compounds, the recipe is telling: the tissue-repair stack BPC-157 + TB-500 is the single most-tested blend in the archive — 118 certificates — dwarfing every GLP-1 combination (the entire grey market mustered only fifteen blend certificates containing a weight-loss drug). The much-hyped CagriSema pairing is still a canary, not a wave: ten certificates, first seen June 2024. And blends inherit the sterility blind spot from the section above — barely 2% of blend certificates carry a purity number at all, so a stacked vial is the least-verified product on the menu.

How to read a certificate of analysis

For anyone who actually relies on these documents:

• Read both numbers. Purity and dose are different measurements. A 99% purity result says the molecule is clean; it says nothing about whether the vial holds the labeled milligrams. Every one of the 1,121 high-purity shortfalls above came with a reassuring purity figure.
• Risk is compound-specific. Insist on a recent, batch-matched certificate far harder for the older peptides — GHK-Cu, CJC-1295, Melanotan II, Ipamorelin, Tesamorelin, all short 17–26% of the time — than for retatrutide or tirzepatide, short around 8–9%.
• A certificate covers one batch, not a vendor. Quality swings batch to batch even within a single seller; last year's clean result does not describe this year's vial.
• One lab underwrites most of this trust. That concentration is its own structural risk.

For a full walkthrough of the labs, the tests, and the costs, see our Lab Testing 101 guide.

This report documents an unregulated market; it does not endorse taking part in one, and it offers no medical or dosing advice. No vendor here is "safe" — the cleanest sellers are clean in this sample, on the certificates we could see. The molecule has gotten clean. The dose is still a coin flip.

Methods & limitations

About this analysis. Figures come from 22,271 third-party certificates of analysis, queried read-only in June 2026. A certificate of analysis is a lab test a buyer or seller posts to demonstrate a grey-market product's identity, purity, and dose. The majority were run by Janoshik (57.3%); most of the rest pass through one aggregator that republishes several labs under coded names.

1. Discovery bias — the most important caveat. This corpus is not a census of grey-market testing; it is what we could discover. About 36% of it (roughly 8,000 certificates) was reconstructed from public data dumps and archived copies of dead lab links. Pre-2023 "history" is 83–100% reconstructed, so any single year's raw volume mostly reflects our recovery effort, not market activity. We therefore never report absolute growth ("testing grew N-fold") and never report a 2026 decline — recent months are undercounted because we have not finished discovering them. Every multi-year claim here is a within-year proportion, robust because under- and over-discovery move the numerator and denominator together. The GLP-1 surge was separately confirmed to be live, contemporary testing rather than a reconstruction artifact.

2. Two numbers, never conflated. Purity (by HPLC) answers whether it is the right, clean molecule. Dose accuracy answers whether the vial held the labeled milligrams, measured strictly as tested-versus-printed-label — deliberately excluding a separate field that folds in sellers' optional batch and overfill claims and would distort the underdose figures.

3. Field coverage. Purity is present on 16,007 certificates (71.9%); label-deviation on 11,335 (50.9%). Anabolic steroids are mostly dose-tested rather than purity-tested, so steroid purity rests on small samples and is read with caution. Blends have no single label dose and are excluded from per-dose statistics. Date fields are OCR-derived and occasionally malformed; time series use well-formed dates only.

4. Attribution. Per-vendor figures are restricted to the 20,340 certificates whose vendor identity has been reviewed and canonicalized (1,547 remain pending, 384 unidentifiable). Corpus-wide aggregate counts use the full archive and say so. A recent attribution pass also consolidated near-duplicate vendor records — one seller had been split across four — so the vendor rankings here reflect merged, deduplicated identities.

5. Anti-overclaim rule. A single bad certificate is never a fraud claim. A vendor is named for underdosing only on a pattern across five or more distinct batches; "worst-rate" sellers whose rate traces to one batch tested repeatedly are flagged as anomalies and not named. Certificate count is volume, never a quality grade — vendors are ranked by rate, with sample sizes shown. We never extrapolate a market or vendor practice from one anomalous batch — the five zero-milligram "vials of nothing" included.

6. Classification. The compound field is free text and partly OCR-derived; drug-class buckets (peptide, GLP-1, anabolic steroid, SARM, hormone) are first-match heuristics with a 0.3% unclassified residual. Class boundaries are fuzzy; the counts within them are exact.

7. Labs measure dose differently — by about nine points. The dose figures in this report blend two measurement regimes, and they do not agree. Holding the compound constant removes any doubt that this is a population artifact: on retatrutide, the largest single source — a coded aggregator lab we call LabE (n≈1,922) — reads a median −0.7% with an 11.9% underdose rate, while Janoshik reads a median +8.6% with a 1.7% underdose rate. Same molecule, overlapping vendors, a ~9-point gap. This is best read as labs measure differently (vendor overfill culture, differing denominators), not one lab is honest and the other isn't — and because LabE supplies so much of the retatrutide data, it pulls our corpus-wide retatrutide underdose figure upward. Corpus dose statistics are a blend; where it mattered we controlled for it.

8. No seasonality. GLP-1 weight-loss drugs hold a steady share of testing every month — within a tight band (roughly 47–55% across 2025–2026, with no summer dip or January spike). The Great Inversion is a multi-year trend, not a seasonal cycle.

9. Blends are barely tested. Only about 2% of blend certificates carry a purity number, and blends have no single label dose, so they sit outside both the purity and the per-dose statistics here. A stacked vial is the least-verified product class in the corpus.

10. Sterility coverage. Endotoxin (injection-sterility) results exist on only 659 certificates (3.0%); the rest are blank or explicitly n/a. Every endotoxin rate in this report uses those 659 measured certificates as its denominator, never the full archive — a "failure rate" here is among the few who tested, and the dominant finding is that 97% did not test at all. NAD+'s near-total failure is 11 of the 12 NAD+ batches that measured endotoxins, not 11 of every NAD+ vial sold.

Every figure and chart is generated directly from the 22,271-certificate database; queries are reproducible from the analysis appendix.