01
No clean pass? We retest.
If the first lab does not return a clean pass, a second independent ISO 17025 lab retests a sibling aliquot from the same sealed batch. We refuse to publish a damaging result based on a single bad reading.
NON-NEGOTIABLE CORE RULES
Transparency you can audit.
Every score on this site is built from independent lab data on samples we bought blind. If a test is not a clean pass, a second lab retests before anything goes live. A third lab is only used when the retest is also inconclusive. The rules below are what we follow on every result, no matter who is involved or how big they are.
THE NON-NEGOTIABLES
Three rules behind every score.
Break any one of these and the test is invalidated. We re-run the chain on a freshly purchased blind sample. Nothing reaches the public record from a compromised process. The rules are enforced in code, not just promised on a page.
02
Blind-bought. Every time.
Samples are purchased anonymously, shipped to random US addresses, and handled by people the vendor cannot identify. They never know what we are testing or when.
03
Zero vendor override.
Sponsor, affiliate partner, advertiser, premium tier. None of it buys a different result. The verdict is published exactly as the lab returned it. No vendor can buy an edit, a delay, or a deletion.
WHY A RETEST
A single bad reading should never destroy a vendor.
Labs are run by humans and machines. Both occasionally fail. A trace contaminant left from a previous run, a calibration that drifted overnight, an aliquot pulled from the wrong vial. These things happen even in ISO 17025 accredited facilities.
The two failure modes are not symmetric:
-
A false fail would torpedo a legitimate vendor who happened to be tested by a noisy lab on a noisy day. From our own experience, about 5% of first-result lab tests are affected by sample handling, instrument drift, or cross-contamination, even at ISO 17025 facilities. A 1-in-20 chance of destroying an honest vendor on a lab artefact rather than on actual product quality is unacceptable.
-
A false pass from that same 5% error band washes out across many audits. A vendor with real quality problems will fail under random sampling within a few rounds, and one lucky reading gets diluted in the running average. The harm is self-correcting over time, so we accept the smaller risk in exchange for faster publishing.
So a clean pass publishes from a single lab. Anything else, fail or inconclusive, triggers a retest at a second independent lab. The retest result is the verdict. A third lab only enters the picture in the rare case where the second test also cannot return a clear answer.
Retests cost more to produce and take longer to publish. That is the price of being fair to the supplier when the result could damage their business.
CHAIN OF CUSTODY
How a test actually happens.
From the moment a sample is purchased to the moment a report appears on the site. Every step.
-
01
Anonymous blind purchase
A team member orders 2 or more sealed units to a random US address using a clean payment method. No VerifiedPure identifiers on the order. Product arrives, gets photographed, and is split into independent sealed aliquots under a blind code so no lab ever sees the vendor name.
-
02
Lab 1 tests aliquot A
An independent ISO 17025 lab runs the full panel. Identity, purity, amount, endotoxin, heavy metals, microbial, residual solvents. The result comes back under the blind code.
-
03
Clean Pass? Publish.
If the result is a clear Pass on every panel, the raw lab PDF goes live with its SHA-256 hash. No second lab is needed. The score gets added to the vendor running average.
-
04
Not a clean Pass? Retest.
If anything came back as a Fail or as Inconclusive, we do NOT publish yet. Aliquot B is sent to a second independent lab. Different city, different equipment, different chemist. Blind code preserved. Whatever Lab 2 finds is treated as the verdict.
-
05
Lab 2 also inconclusive? Lab 3 decides.
A third lab is only used in the rare case where Lab 2 also cannot return a clear verdict. Aliquot C is sent to Lab 3 as the decider. Vendor is never told. Once a verdict exists, the report publishes unedited with every raw PDF and every SHA-256 hash side by side.
Clean pass publishes from one lab
Anything else triggers a retest
Third lab only if retest is also inconclusive
THE ONE-WAY RATCHET
Once a score is published, there is only one way it changes.
By waiting for the next random audit. Nothing else moves it. Not money, not pressure, not time, not apologies.
Worked example. One bad test, 100 audits later.
A vendor fails their first audit. Heavy metals over spec. Here is what their public score looks like as more random audits arrive.
| Tests in record | What happened | Published score | Note |
|---|---|---|---|
| 1 | 1st audit. Heavy metals fail. | 2.0 /10 | Below minimum sample threshold. No public score yet. |
| 3 | 3 tests in. 1 fail + 2 strong pass. | 7.4 /10 | Minimum threshold met. Public score visible. The fail is still in the average. |
| 10 | 10 tests. 1 historical fail. | 9.0 /10 | Bad test never deleted. Its weight is now 1/10 instead of 1/3. |
| 50 | 50 random audits. Same 1 fail. | 9.6 /10 | The fail still counts. Its weight is now 1/50. |
| 100 | 100 audits. | 9.8 /10 | Average converges on actual quality. No shortcuts taken. |
The first fail is still in the average at test 100. Its weight is now
1/100. Bad history never disappears, but consistent good work dilutes it. Bad work concentrates it. This is what fair looks like.
THE MATH
The formula. Deterministic. Public. Hash-anchored.
The same inputs always produce the same number. No human input touches the math. The formula source code is open, versioned, and cryptographically anchored.
Per-test score (0 to 10)
Each published lab test produces a score from 0 to 10. It is a weighted sum of seven criteria.
Identity verified
Pass = 10. Fail = 0.
Purity vs vendor claim
Linear. 10 points at 100% purity. 0 at 80%. Below 80% = 0.
Amount accuracy
10 points at ±2% of claim. 0 at ±20% or worse. Linear in between.
Endotoxin within spec
Pass = 10. Fail = 0.
Heavy metals within spec
Pass = 10. Fail = 0.
Microbial within spec
Pass = 10. Fail = 0.
Residual solvents in spec
Pass = 10. Fail = 0.
Aggregate score
vendor_score = arithmetic_mean(all published per-test scores)
A vendor needs at least 3 published tests before a public score is displayed. Below that threshold the listing shows "Sample size too small" and the vendor sits at the bottom of the rankings.
score-engine.php · public · open-source
// Per-test score, 0 to 10. Pure function. Deterministic. function per_test_score(array $r): float { $identity = $r['identity_pass'] ? 10 : 0; $purity = max(0, ($r['purity_pct'] - 80) / 20 * 10); $amount = max(0, 10 - max(0, $r['amount_dev_pct'] - 2) / 18 * 10); $endo = $r['endotoxin_pass'] ? 10 : 0; $metals = $r['heavy_metals_pass'] ? 10 : 0; $micro = $r['microbial_pass'] ? 10 : 0; $solv = $r['solvents_pass'] ? 10 : 0; return round( $identity * 0.20 + $purity * 0.25 + $amount * 0.20 + $endo * 0.10 + $metals * 0.10 + $micro * 0.10 + $solv * 0.05, 2); } // Vendor score = arithmetic mean of all published per-test scores. function vendor_score(array $tests): ?float { if (count($tests) < 3) return null; // not enough yet $sum = 0.0; foreach ($tests as $t) $sum += per_test_score($t); return round($sum / count($tests), 2); }
Zero deviation. Reproducible by anyone.
Download the raw lab PDFs from any vendor profile. Read off identity, purity, amount, contaminant verdicts. Plug them into the published formula. The number you get back has to match the number we show. If it does not, that is a bug worth a public dispute and we will fix it on the record.
AUDIT CADENCE
The vendor cannot predict when, what, or where.
The next random audit is the only lever a vendor can pull on their own score, but they cannot pull it themselves. We choose the timing.
Time since last audit
Vendors not tested recently move up the queue. A long quiet period guarantees the next purchase.
Audience demand
Compounds and vendors that researchers are asking about get tested sooner. Search and click signals feed the cadence.
Dispute volume
A vendor with active disputes or anomalous prior results is sampled more frequently and from more batches.
Sample-size threshold
Vendors with fewer than 10 published tests are prioritized so the public score has statistical weight.
What we never disclose: the next purchase date, the next product, the next lab, the shipping address used, the test runner, or the alias on the order. Vendors learn about an audit when the report goes live.
INPUTS TO THE SCORE
Where the score actually comes from.
The score is decided by the lab data, full stop. We keep a public list of what does and does not reach the scoring engine, so anyone can audit it line by line.
What the score is built from
- What the independent lab actually measured
- Whether the confirming lab (when a fail required one) reached the same verdict
- Whether contaminant thresholds were met
- Whether the chain-of-custody was sealed end to end
- Whether the raw report PDF hash matches what we received
- How many independent tests have been published
What never enters the calculation
- ×Affiliate revenue
- ×Sponsor or "Featured" placement
- ×Vendor program tier (Free, Listed, Partner)
- ×Coupon code partnerships
- ×Ad spend on the directory
- ×Vendor dashboard subscription
- ×Personal relationships with the lab or vendor
- ×Country of origin
- ×A vendor sending us "samples for review"
- ×Vendor anger, threats, or legal pressure
The right-hand list is enforced in code by the verifiedpure-core plugin, so a developer would have to touch the score engine in version control to change what counts. Any place on the site that hides the source of a recommendation is a bug worth reporting.
OUR PLEDGE
Seven things that will never happen on this site.
- ×No vendor sees a result before publication.
- ×No score is influenced by money, friendship, or pressure.
- ×No published test is ever deleted, hidden, or reweighted.
- ×No retest is granted to "improve" a score.
- ×No vendor can pick which lab tests their product.
- ×No data is hidden. Raw PDFs, lab signatures, and SHA-256 hashes are public.
- ×No relationship, past or present, overrides a failing panel.
If we ever slip, the original record stays public and so does the correction. The history of what changed is part of the data.
FAQ
Common questions, plain answers.
Why does a passing result publish from a single lab?
A false pass is a small risk. If a lab accidentally returns one inflated reading, it sits in the running average and gets diluted by every future random audit. A vendor with real quality problems will fail under random sampling within a few rounds. A false fail however cannot be undone in the public eye, so anything other than a clean pass gets the retest.
What happens when a lab returns a fail or an inconclusive result?
Nothing is published yet. A sibling aliquot from the same sealed batch is sent to a second independent ISO 17025 lab in a different city. Whatever the second lab finds becomes the verdict. A third lab is only used in the rare case where the second test is also inconclusive. The vendor is never told the retest happened.
Can a vendor pay to retest a failing product?
No. Paid retests do not exist on VerifiedPure. We only retest when our own internal QC flags a chain-of-custody issue. The retest is run on freshly purchased blind samples, never on samples the vendor sent us.
Can a vendor see results before they are made public?
No. Vendors are notified at the moment a report goes live. They have no opportunity to "correct" or suppress findings prior to publication. Vendor responses are recorded post-publication, in the public response log.
Can a vendor improve their score?
Yes. The only path is to wait for the next random audit and pass it. Each new test gets added to the running average. A bad test from years ago never disappears. Its weight just dilutes as the body of audited tests grows. This rewards vendors who can pass 100 random tests, not vendors who can pass one cherry-picked test.
How do I know the formula was not bent to favor a sponsor?
The formula is published with a SHA-256 hash of the source code. Each vendor profile shows the version of the formula in effect. You can download the raw lab PDFs, plug the values into the formula, and recompute the score. If the recomputed score disagrees with the published score, that is a bug we want reported.
What if a vendor refuses to send invoices or proof of manufacturing?
Their listing remains in the directory but with a clearly labelled "Origin Unknown" badge. Their score is still computed from independent lab tests on blind-bought samples. We do not need vendor cooperation to test a vendor.
Who pays for all this testing?
A mix of revenue streams independent of the score: data API access for developers, vendor dashboard tooling, and clearly labelled outlet placements. Every commercial relationship is listed publicly on the Ad Transparency Register. None of it touches the scoring engine. The engine reads the lab data and computes the number. That is enforced in version-controlled code, not policy.
Found something off? Tell us.
If a test result looks wrong, a page hides where a link came from, or the rules above are not being followed somewhere, we want it on the record. Open a public report and the whole thread stays visible.