What 1,816 Kovera Labs COAs reveal about the US peptide market

Short version. Kovera Labs is a US-based independent analytical lab that runs HPLC-MS on research-peptide samples and publishes every certificate to a public verifier at koveralabs.com. We mirrored their entire public corpus. The dataset is 1,816 published COAs across 146 different vendor-clients between January and May 2026, with a median purity of 99.907% and an explicit failure rate of 0.8%. The lab publishes failures alongside passes, which is the structural property that distinguishes a credible third-party operation from a vendor-controlled testing pipeline.

The longer version covers what the corpus shows about the lab itself, which vendors come out clean, which vendors come out flagged, and what a buyer can and cannot conclude from any of it.

What we built

Every Kovera COA carries a per-record verifier URL of the form koveralabs.com/verify?code=KVR-2026-XXXXXX. The verifier returns the headline assay data directly: purity, identity confirmation, lot number, sterility, endotoxin, heavy metals, and pass/fail per assay. Unlike Janoshik (where we extract purity numbers from COA images using vision OCR) and unlike Ascension and Swiss Chems (where we extract from vendor-self-published images), Kovera surfaces structured COA data per record.

That makes Kovera the first source in our pipeline where extraction costs us nothing. No OCR, no LLM transcription, no per-record API spend. The deterministic path covers the 16-field COA schema directly from the per-record verifier response, with a pdftotext pass on the COA PDF to lift the Client: field (the one piece of metadata not on the verifier face).

The full pipeline build is documented internally. What matters editorially: every number in this article was extracted directly from Kovera's published data, not from a vendor's republication of it.

The headline numbers

The five-month publication window (January 11 to May 25, 2026) contains:

• 1,816 published COAs in the registry. 1,749 are full primary-product COAs; the remaining 67 are standalone supplementary tests (heavy metals, endotoxin, bac water, sterility).
• 146 unique vendor-clients with at least one published Kovera COA.
• Median purity of 99.907% across the 1,379 records reporting a purity figure. p10 is 99.629%, p90 is 99.976%.
• 15 explicit purity failures (0.8% of result-bearing records).
• 7 identity-mismatch records across 5 different vendors.
• 3 catastrophic 0% purity records where the lab could not detect the labeled compound at all.

The median tells the most important story. Typical Kovera-tested product comes back above 99.9% pure. Anything below 99% on a Kovera report is unusual, and anything below 95% is rare enough that we can list every record below that threshold by name (we do, below).

The 0.8% explicit failure rate is also the structurally important number. It is non-zero. A lab that publishes 0% failures is not credibly publishing all results.

What this proves about Kovera as a lab

We use a four-question framework to evaluate any third-party lab as a credible source. It comes out of our Janoshik audit and the same four questions apply to Kovera.

1. Is the corpus broad across vendors? Yes. 146 distinct vendor-clients across 1,816 records means no single vendor shapes more than 6% of the published pool (Ion Peptide tops the leaderboard at 105 records). A captive lab serving a single customer has a flat distribution against one client; Kovera's distribution is a long tail.

2. Are failures visible in the public corpus? Yes. 15 explicit purity failures, 7 identity-mismatch records, and 3 zero-purity records all sit in the public verifier alongside the passes. No retraction, no quiet removal.

3. Does the verifier resolve to lab-controlled records? Yes. A Kovera verification URL (koveralabs.com/verify?code=KVR-...) returns data from Kovera's servers, not from the vendor's CMS. A vendor cannot edit the underlying record.

4. Is the methodology consistent? Yes. The structured results blob carries the same assay schema across vendors. There is no evidence of vendor-specific shortcuts in the published corpus.

Two third-party labs (Janoshik in the EU, Kovera in the US) clearing the same four checks is the structural argument for what "legit" means in this market. It is not a claim that every COA is meaningful. It is a claim that the lab is publishing what it finds.

The vendor leaderboard

30 vendors have at least 20 published Kovera COAs. The table below shows the heavy publishers ranked by record count, with three flag columns: fail = lab marked the record passed: false, id-mis = identity confirmation flagged, low-p = purity below 95%.

Three observations on the distribution:

Most heavy publishers come out clean. Ion Peptide at 105 records and zero flags is the top of the leaderboard, and it has plenty of company. Zenergy (58), Coffee and Peppers (54), Mile High Compounds (52), BioEdge Research Labs (40), Fixion Fuel (40) all carry zero flags across their published volume. That's the shape of a vendor pool where typical practice clears the bar.

The flag clusters concentrate in two vendors. Of the 30 heavy publishers, only Instant Peptides (5 fails of 72) and Glacier Aminos (3 fails of 77) carry concentrated quality issues. Neither is rare enough in the raw vendor pool to be ignored, and both have published enough records that the failure rate is not a one-off.

Personal-email "vendors" appear in the data. Names like jakestudzinski@studzpeptides.com and support@purasynth.com show up as Kovera clients. These are real vendors using personal-identity submission, not corporate identities. Worth flagging in any subsequent audit profile because it changes the accountability picture.

Where the failures live

Three failure clusters deserve a closer reading. The lab caught them. Vendors who routinely submit to Kovera and accept publication of the results are essentially guaranteeing this evidence stays in the public record.

The 0% purity records

Three records came back at 0.0% purity. At that magnitude, the lab is not measuring an impure sample - it is reporting that the labeled compound could not be detected at all.

GLP-We1 published two consecutive 0% purity COAs on different active ingredients. That is past the threshold for a single batch-quality anomaly and into pattern territory. A vendor whose lab cannot detect the labeled compound in successive submissions is selling something other than what the label says, and we would not buy from them.

Sub-95% purity records

Two records came back well below industry baseline but above zero:

The 46.85% BPC-157 from Axon Peptide is alarming in absolute terms. A pure-versus-impure peptide does not run that low; that is a different molecule with some BPC-157 contamination, or BPC-157 with massive contamination. Either way the vial does not contain what the label says. This is a single record, so we are not pattern-flagging Axon Peptide yet. A second sub-spec record would change that.

The 72.85% Epithalon from Peptide Partners is poor but consistent with synthesis-batch quality issues at the manufacturer level. Worth noting; not a fraud signal on its own.

Identity-mismatch clusters

identityConfirmed: false means the molecule the lab found did not match the reference compound the vendor said they sent. There are 7 such records across the corpus.

Two of those clusters are structurally different and deserve different reads:

Instant Peptides carries identity-mismatch records on legitimate single-molecule peptides where the reference standard is well-defined. NAD+ and Tesamorelin both have known molecular weights and the lab is essentially saying "the molecule we found does not match the reference." That is the same signature you would see on counterfeit or grossly under-spec material. Three such records out of 72 (4.2%) is well above the baseline of zero we see on every other heavy publisher except Glacier Aminos.

Glacier Aminos carries identity-mismatch records on proprietary blends with vendor-coined names ("GLA-3 RT", "MTP-31", "Glow"). Identity confirmation on a blend is structurally harder because there is no single reference compound to compare against. The lab flagging identity on these is not the same kind of evidence as the Instant Peptides flags. It is more consistent with "the published spec sheet does not match what we received in the vial" than with "the vial contains counterfeit material." Worth investigating whether Glacier Aminos has non-blend single-peptide products that pass cleanly.

The Klow / Glow stack records

Two records validate the read we already published in the KLOW stack literature article:

Both stacks carry high single-peak purity numbers and identity-mismatch flags simultaneously. That combination is the fingerprint of "we measured purity of one of the four claimed ingredients, and even that one did not match the reference." When a vendor publishes a single purity figure for a multi-component blend, they are publishing the purity of one peak - typically the cheapest or most stable component. The lab cannot verify the identity of the full stack because the stack as labeled does not exist in any reference database.

The buyer takeaway: 99.9% on a blended-product COA does not mean 99.9% of the labeled stack. It means 99.9% of whatever the lab measured. The two only become the same number when there is one ingredient in the vial.

What the data does NOT show

Three limitations worth surfacing before drawing conclusions.

Self-selection bias. Kovera tests what vendors submit. A vendor that runs internal pre-tests and only submits favorable batches to Kovera does not produce failures in the public corpus. The 0.8% explicit fail rate is a floor, not a ceiling.

Five-month window. Kovera began publishing in January 2026. The public corpus has not yet seen a full year of vendor behavior. Some seasonal patterns (Q4 manufacturing rushes, summer heat-stability failures) cannot be assessed from this window. Re-run this analysis in November 2026 for a more representative read.

No cross-source dedupe yet. A batch tested by Kovera and re-tested by Janoshik is not currently joined in our index. We do not yet know which vendors are double-testing across both labs. That joining work is in the queue and would produce a stronger read on which vendors actually publish their internal failures versus which only publish to one lab at a time.

What this means for buyers

Three operating rules consistent with the Kovera data and the Janoshik data combined.

Prefer vendors with multi-record published histories. A vendor with one Kovera COA on one batch tells you something. A vendor with 50 Kovera COAs across multiple products and dates tells you something stronger: they are routinely submitting to a lab that publishes failures, and the failures they have published are within the industry-baseline rate. Ion Peptide, Zenergy, Coffee and Peppers, Mile High Compounds, and BioEdge fit this pattern in the current corpus.

Treat identity-mismatch flags as binary. Sub-spec purity is a gradient. Identity-mismatch is not. A flagged identity record means the lab could not confirm the molecule matches the reference. For single-ingredient peptides this is the strongest negative signal in the dataset and the one a buyer should weight most.

Match the lot to the COA. Every Kovera COA carries a lot number. The vial in your hand should also carry a lot number. They should match. A vendor publishing a Kovera COA from a clean batch and shipping you a different batch with the same SKU label is the classic substitution failure mode, and the lot match is the only buyer-side defense against it.

What this is not

This analysis does not assess Kovera's analytical methodology. We have not audited their HPLC columns, mass-spec calibration, or chain-of-custody handling. The structural property we have validated is publication discipline (failures visible, multi-vendor corpus, verifier outside vendor control). The chemistry on any individual COA is what Kovera says it is, the same way the chemistry on any individual Janoshik COA is what Janoshik says it is.

It also does not replace our first-hand test orders. The 1,816 records measure what vendors choose to submit to Kovera. The vendors that route batches around Kovera entirely are by definition not in this corpus. First-hand audits are the only way to test what a vendor ships when they do not know they are being measured.

Related reading

• What 7,164 Janoshik tests reveal about peptide vendor purity - the analog piece for the EU multi-vendor third-party lab corpus.
• Is Janoshik testing legit? - the four-question framework applied to Janoshik, and the article we extended to acknowledge Kovera as a second multi-vendor third-party lab passing the same checks.
• How peptide vendors fail: a taxonomy - the nine failure modes the Kovera data confirms, plus the captive-lab pattern that Kovera explicitly does not match.
• KLOW stack literature read - the article these two stack-product identity-mismatch records (KVR-2026-5953A8 and KVR-2026-50085D) now appear in as cited evidence.
• Our methodology - what audits look like when first-hand test orders are added on top of public COA mirroring.

Sources

• Kovera Labs public verifier - the source-of-truth verifier audited for this piece
• Kovera Labs verifier endpoint - the lookup pattern: koveralabs.com/#/verify?code=KVR-2026-XXXXXX
• Internal: 1,816-record extraction corpus at data/kovera/ (May 2026 mirror, per-record verifier extraction)
• Internal: What 7,164 Janoshik tests reveal
• Internal: Is Janoshik testing legit?
• Internal: How peptide vendors fail: a taxonomy