01Methodology

How a vendor is checked

Every certificate a vendor publishes is looked up at the laboratory that issued it, using the identifiers printed on it.

Outcomes: 7
Score categories: 7
Laboratories: 33
Automated checkers: 19

01Outcomes

What each result means

Seven answers, fixed. Each one is a statement about a document, never about a company.

Authentic
The laboratory holds it and its own copy names this vendor.
Misattributed
The laboratory holds it and issued it to another company.
Edited
The laboratory holds it and the vendor's copy differs.
Not found
The laboratory has no report under this number.
Unverifiable
There is no public laboratory original: no portal, a public search that does not list it, or an original that requires identity verification.
Cannot verify
A documented omission prevents independent verification: no independent issuing laboratory and verification route are identified. These findings count against the vendor.
Pending
Looked up, comparison still running.

02The rule

The rule that fails a vendor

One report is enough to fail a vendor.

Not foundMisattributedEditedCannot verify

Any one of these four fails the vendor. The count does not soften it and no other category outvotes it.
Cannot verify is a certificate whose saved source names no independent issuing laboratory or verification route. The finding does not establish fabrication. Laboratories that publish no public verifier (Krause, Horizon certificates without a report number) are grey Unverifiable instead.
A document presented as a COA also fails these checks when review of the document and its source identifies no independent issuing laboratory or verification route. The saved source evidence must show this omission. A supplier’s own quality-control statement is not independent laboratory verification. This finding does not establish fabrication or an accreditation status.
File format does not decide the result. A text file, PDF or image is assessed against the same requirement: an identified independent laboratory and a way to confirm the report with it.
A documented review that finds no publicly posted COAs also fails the public-COA requirement. The record lists the pages reviewed, date and observations. It carries no laboratory verification results or numerical score, and does not establish that testing never occurred. A crawler that cannot access the storefront is not evidence that public COAs are absent.
Unverifiable does not fail a vendor and does not zero authenticity. It is explained on the record. A vendor with nothing confirmed is reported unverifiable, never verified. Laboratories with no public verifier (Krause, Horizon certificates without a report number) use this grey outcome, as do originals that require identity verification before they open (BTLabs).
When every published certificate is Unverifiable, the vendor verdict is unverifiable, not failed. Those suppliers have their own rankings category, separate from Failed and Verified, and they do not set market prices.

The homepage failure rate counts published, completed audits with a verified, failed or unverifiable verdict. It divides failed or unverifiable suppliers by that total and rounds to the nearest whole percent. Pending verdicts are excluded. A grey Unverifiable card still counts on the failed side of this figure. The percentage appears once ten suppliers qualify; below that, we show counts. It describes this register, not an estimate of the entire market.

03Score

7 categories, 100 points

Each category is scored 0 to 100 and multiplied by its weight. The weights are fixed in code and every score on the site is computed from this table. A category that cannot be measured is left out of the total rather than counted as zero, and a record carrying one stays a draft until a person settles it.

#CategoryPointsMeasures, then the bands

01Report authenticity20Can the issuing laboratory independently verify the reports the vendor publishes?Scored first. An edited copy or a report issued to another company scores 0/20 and stops the rest of the rubric. A portal miss from a laboratory that should hold the record still accuses and also scores 0. Grey Unverifiable rows — a laboratory with no public portal, a public search that does not list the certificate, or an original that requires identity verification / a login — do not accuse. When nothing accuses, the line is the confirmed share of non-accusing reports, round(100 × authentic / (authentic + unverifiable)), mapped onto 20 points, so 12 confirmed beside 6 Unverifiable is 12/18 → 67 → 13/20. The published count in the sentence is still every report the storefront publishes, so 12 of 19, never 12 of 12.

02Testing coverage10What share of listed products, and of each size sold, carries a published report?Measured against sizes rather than products, because a vendor that tests one size of a peptide and sells four has tested a quarter of what it sells. One report against a large catalogue scores near zero.

03Report recency10How recently was each published report issued by the laboratory?Full marks only when every report is inside 3 months. Beyond that a report scores 80 to 6 months, 55 to a year, 25 to two years, and 5 after that, and the category is the average across reports.

04Measured purity10What purity did the laboratory actually report, averaged across confirmed reports?Two points for each 0.1% of average purity, reaching 10 at 99.5% and nothing at 99.0%. A tenth not reached is a tenth not earned, so 99.26% is read as 99.2% and scores 4. The line is then capped by how much of the catalogue those figures cover: half or more scores up to 10, a quarter up to 8, a tenth up to 6, under a tenth up to 4, and a bare fraction up to 2. One certificate cannot speak for two hundred products.

05Legal compliance10Whether the storefront and its paid advertising stay inside research-use language.The former storefront-compliance and advertising-compliance lines, scored together. Marketing these compounds for human consumption on the site or in ads scores 0. A research-use gate, labelling on product pages, copy free of mixing instructions, and a way to reach the company still count on the storefront half. Ads are checked against Meta's public Ad Library: none, or research-use only, is a clean advertising half.

06Pricing20Price per milligram against the market median for the same compounds.0.6x the market median or below scores 100, market rate scores 70, 1.5x scores 35, 2x and above scores 0. The market table is rebuilt from catalogues that can still set a price: failed and all-Unverifiable vendors are excluded, so it moves as the register grows. A compound needs prices from five separate vendors before it is published as a rate, and a vendor selling only compounds below that floor is not scored on price.

07Storefront20Dispatch, whether the shop can be moved through, the range on the shelf, and what the vendor discloses.Shipping, navigation, product selection and transparency, scored as one shop line. Dispatch under 2 business days, tracking and cold-chain handling count. Navigation is whether a catalogue, policy pages and a contact route can actually be reached; page load is not timed by this register. Selection is a log curve on the catalogue size. Contact routes, policies and batch numbers on reports make up the rest.

A failed vendor is still scored on authenticity only. An edited copy or a report issued to another company scores 0/20 and stops the rest of the rubric. A portal miss from a laboratory that should hold the record still accuses. Grey Unverifiable reports — no public portal, or a public search that does not list the certificate — are explained and do not themselves zero this line; when nothing accuses, authenticity is the confirmed share of those non-accusing reports, mapped onto 20 points.
Pricing is measured against the register itself: a median price per milligram per compound, rebuilt from every audited vendor whose certificates did not fail, and published only once five separate vendors sell the compound.
Storefront and advertising compliance are scored together as legal compliance. Dispatch, navigation, product selection and disclosures are scored together as the storefront line. Page load is not timed.
Every published vendor is audited again once a month. The record carries the date it last ran.
A score summarises these categories. It is not a measure of safety, legality or fitness for any purpose. See the disclaimer.

04Laboratories

Where the lookups run

33 laboratories are on the register. 19 have an automated checker in our code; a report from one of the rest is opened at whatever address that laboratory publishes. Each one, its verification route and its lookup address are on the laboratories page.

A laboratory is recorded here when its verifier requires a report number and certificates it issues are found that print none. The record carries the date it was made, and a certificate from that laboratory that prints no report number is Unverifiable (grey), not Failed. Recorded now: Horizon Analytical (2026-09-11).

05Standing

What a record stands on

A finding describes the certificates a vendor published on the dates listed. Each carries the portal that answered, the identifiers used, the time in UTC, and an exhibit with a SHA-256 hash, so a reader can repeat the lookup.
A vendor that removes or replaces a failed report can ask for a new check through the response form. The earlier finding stays published with its date and evidence.
Errors are handled under the corrections policy.