Between 1 and 2 August 2026 we re-fetched all 1,828 Kovera Labs certificates in our mirror. 1,572 were unchanged, 185 had values filled in where we previously held nulls, 3 had been unpublished, and 57 had been revised under the same certificate number and the same URL. Of those 57, 20 moved an actual purity or content value. Four purity moves exceeded 0.3 percentage points, and none crossed the 98% pass threshold in either direction, so no pass or fail verdict changed. The practical conclusion is narrow and important: a Kovera COA number is not a fixed citation the way a Janoshik test ID is, so record the date you retrieved one.
We assumed a certificate number was a permanent address. It is not. Re-running every one of the 1,828 Kovera Labs certificates in our mirror found 57 that had been re-certified under the same number, at the same URL, with no notice. Twenty of them carry a purity or content figure different from the one we recorded.
The honest headline is the second sentence, though: none of the changes crossed the pass line. Nothing flipped from pass to fail or back. This is not a story about a lab hiding failures. It is a story about what a COA reference number actually is, which turns out to be a pointer to the current version of a document rather than to a fixed historical record.
What we ran, and what came back
Our mirror holds every certificate Kovera had published as of a crawl in late May 2026, plus records added since. On 1-2 August 2026 we re-fetched all of them through Kovera's public per-record verifier and compared each field against what we had stored.
| Outcome | Records | Meaning |
|---|---|---|
| Unchanged | 1,572 | no value moved |
| Backfill | 185 | we held null, the lab now supplies a value |
| Revision | 57 | a value we held was replaced, or the record was re-published |
| Gone | 3 | the verifier returned nothing |
| Failed | 0 | - |
| Total | 1,828 |
The 185 backfills are our problem rather than Kovera's. Those are records where our earlier extraction stored a null because the structured data was not populated at the time, and the figures were available all along. Our published index was incomplete, not wrong. Fixing the extraction lifted purity coverage from 1,564 records to 1,741.
The 57 revisions are the interesting set.
What actually changed in the 57
Not all 57 moved a number. In 37 of them the server-side modification timestamp advanced with no visible change to purity or content, which is consistent with a record being re-published rather than re-measured.
Twenty moved a real value:
| Change type | Count |
|---|---|
| Purity value replaced | 19 |
| Net content value replaced | 8 |
| Purity move of 0.3 percentage points or more | 4 |
| Purity move crossing the 98% pass threshold | 0 |
The largest moves:
| Certificate | Compound | Purity | Net content |
|---|---|---|---|
| KVR-2026-070EFA | hGH w/ dimer | 97.83% → 94.83% | · → 8.48 mg |
| KVR-2026-AEC9DB | PT-141 | 98.55% → 99.53% | 11.06 → 11.33 mg |
| KVR-2026-9F405D | BPC-157 | 99.36% → 99.77% | 5.19 → 5.67 mg |
| KVR-2026-129E57 | NAD+ | 99.63% → 99.40% | 508.65 → 543.18 mg |
| KVR-2026-019FBF | Semax | 99.51% → 99.67% | 11.38 → 5.59 mg |
| KVR-2026-D560C6 | IGF-1 LR3 | 99.92% → 99.52% | 1.14 → 1.16 mg |
The Semax record is the one worth staring at. Its purity figure went up by 0.16 points while its measured content fell by more than half, from 11.38 mg to 5.59 mg. Anyone tracking that certificate on purity alone would have seen a marginal improvement and missed a vial that lost half its peptide. Purity and content are different measurements and they can move in opposite directions.
_in_CAFIA_laboratory,_Czech_Republic.jpg){kind=link}
The revisions are concentrated, not diffuse
This is the part that was not obvious until we grouped the data, and it changes how the finding should be read.
Our corpus spans 146 vendor-clients. The 57 revisions touch 14 of them.
| Vendor on report | Revisions | Of which moved a value |
|---|---|---|
| Mile High Compounds | 19 | 5 |
| Evolabs Research | 7 | 7 |
| Axon Peptide | 7 | 0 |
| Glacier Aminos | 5 | 1 |
| Peptide Labs | 5 | 0 |
| RedlinePeppers | 3 | 0 |
| BioEdge Research Labs | 3 | 3 |
| 7 other vendors | 8 | 4 |
The revisions also cluster by date rather than trickling in. Eleven landed on 27 May 2026, seven on 31 May, seven on 9 June, six on 20 June.
Two readings fit this and we cannot distinguish between them from outside. A batch of records being re-published on a single day looks like an operational event at the lab, such as a re-extraction or a data migration. It could equally be a client requesting re-analysis of a group of samples. What we can say is that this is not gradual drift across the whole corpus, and that every value-moving revision for Evolabs Research landed on one day, 30 May 2026.
We are not drawing an integrity conclusion from that. We are recording it, because the pattern is the kind of thing that only becomes interpretable with a second observation, and now there is a baseline to compare against.
Bulk discovery is closed
A second change surfaced during the same exercise. Kovera's public listing endpoint, which previously let anyone enumerate the full published corpus, now returns an empty set. The per-record verifier still works normally for any certificate number you already hold.
The practical effect is that certificates published after our May crawl cannot be found by listing. We recovered part of that by a different route: vendors who submit samples to Kovera publish the resulting PDFs on their own sites, and every PDF carries its certificate code. Harvesting codes from vendors' own published files and confirming each through Kovera's public verifier found 11 certificates we did not have on a first pass against one vendor, taking the registry to 1,828.
We take no position on why the listing endpoint changed. It is Kovera's data and closing bulk access to it is their call. The consequence for anyone doing this kind of work is simply that corpus-level analysis of this lab now requires vendor-by-vendor discovery rather than a single query.
What this changes about citing a COA
A Janoshik test ID behaves the way people assume lab references behave: it points at one report. A Kovera certificate number points at the current version of a record that can be re-issued underneath it. Both are legitimate laboratory practices. Only one of them is safe to screenshot.
So the convention we are adopting across this site, starting with this piece:
Every COA we cite carries a retrieval date. Not just the certificate number and the lab, but the date we last pulled the record. A figure without a retrieval date is a figure of unknown age.
For a reader, the same rule collapses into three steps.
Re-check before you buy, not before you bookmark. A certificate you verified in May is evidence about May. If the purchase is happening now, pull the record now. Our COA verifier queries the issuing lab rather than the vendor's copy.
Read the content field, not only the purity field. The Semax record above is the argument. Purity can improve while content collapses.
Treat a vendor-hosted PDF as a starting point, never as the evidence. That holds regardless of revisions, and the reason it holds was established in a federal courtroom this year, in the Paradigm Peptides forged-certificate case.
What this does not show
Worth being explicit, because a finding like this is easy to over-read.
It does not show Kovera concealing failures. Not one revision moved a result across the 98% pass line, in either direction, and the lab continues to publish outright failures including 0% purity records.
It does not show the revised figures are wrong. A laboratory re-analysing a sample and updating a result is doing its job. The gap is in versioning and notification, not in measurement.
It does not tell us how common this is at other labs. We hold comparable mirrors for Janoshik, Liquilabs, MZ Biolabs and others, and re-running those the same way is the obvious next step. Until we do, the honest statement is that we found this at one lab because we looked at one lab.
Related reading
- What 1,816 Kovera COAs reveal - the corpus analysis this refresh was run against.
- Is Kovera Labs legit - the four-check framework applied to the lab itself.
- Kovera vendor leaderboard 2026 - per-vendor standing in the same corpus.
- How to verify a Janoshik COA - the manual check, and why a lab-hosted record beats a vendor-hosted PDF.
- Forged COAs and a 70-month sentence - the failure mode that makes vendor-hosted certificates worthless.
- How peptide vendors fail: a taxonomy - where stale and fabricated certificates sit among the failure modes.
Sources
- Kovera Labs public verifier - every figure above is derived from our mirror of this data, refreshed 1-2 August 2026
- Internal: Kovera corpus analysis, is Kovera Labs legit, Kovera vendor leaderboard
- Internal: our methodology - how the mirrors are built and refreshed
Frequently asked
Do peptide labs change COAs after publishing them?
Kovera Labs does. Re-fetching all 1,828 certificates in our mirror on 1-2 August 2026 found 57 that had been re-certified under the same certificate number and the same URL, with 20 of those carrying a changed purity or net-content value. The document at a given reference number is current rather than permanent.
Did any Kovera COA change from a pass to a fail?
No. Nineteen certificates moved a purity value and eight moved a net-content value, but not one crossed the 98% pass threshold in either direction. The largest purity move, from 97.83% to 94.83% on an hGH product, started and finished below the line. No verdict flipped.
What was the biggest change in a revised Kovera COA?
By purity, an hGH-with-dimer certificate moved from 97.83% to 94.83%. By content, a Semax certificate moved from 11.38 mg to 5.59 mg, a drop of more than half, while its purity figure rose slightly. That combination is the clearest illustration that purity and content are separate measurements.
Can you still look up any Kovera COA?
Yes, if you already have the certificate number. Kovera's per-record verify endpoint still returns full data for any code you hold. What closed is bulk listing: the public endpoint that let anyone enumerate the whole published corpus now returns an empty set, so new certificates can no longer be discovered that way.
How should I cite a peptide COA?
Record the certificate number, the issuing lab, and the date you retrieved it. The retrieval date is the part most people skip and the part that matters, because a certificate that read 99.5% in May may read differently today. For any purchase decision, re-check the record rather than relying on a screenshot.
Does this mean Kovera Labs is not trustworthy?
Not on this evidence. Revising a certificate after re-analysis is normal laboratory practice, and none of the revisions we found moved a result across the pass line or hid a failure. What the finding shows is that the certificate number is not a permanent identifier, which is a citation problem rather than an integrity finding.