The supplier leaderboard is driven by a single composite score. This page is the formula, line by line, with the weights published and the edge cases named. Anyone — a vendor, a researcher, a competitor, or a regulator — should be able to read this page and explain why a specific ranking moved when it moved.
Click any underlined term (e.g. , , , , ) for a plain-English explanation.
Most existing peptide rankings do one of two things. Some publish a single pass/fail per test with no aggregation, which makes the data honest but useless for picking a supplier. Others publish an aggregate that reflects editorial preference without showing the math, which reverses the problem. We try to do both — publish every underlying assay at /assays, and publish the exact formula that rolls them into the leaderboard number.
The composite score sits between 0 and 100. Every input is bounded in [0, 1] and multiplied by its weight; the sum is multiplied by 100 to produce the public number. Grades are assigned on fixed thresholds: A ≥ 85, B = 70–85, C < 70. Grades are a UI convenience; the score is the data.
The single most important axis. A peptide that chromatographs cleanly but does not mass-spec to the expected molecular weight is the wrong molecule — full stop. We weight identity the highest because it is the cheapest check for a vendor to pass (or fail) and the most brand-damaging when wrong.
Area-percent at 214 nm, evaluated against the same 98% research-grade threshold regardless of what the vendor publishes. Degraded or under-synthesized material with significant related-substance peaks fails here. The weight is slightly below identity because identity is categorical while purity is a gradient.
How tightly purity and identity cluster across multiple batches of the same peptide from the same vendor. A vendor with a single perfect test is not the same as a vendor whose five most recent tests of the same peptide all passed within measurement tolerance. We compute a standard deviation on per-batch purity and invert it — tighter cluster, higher score.
Older test results tell us less about today's batches than recent ones. We apply an exponential decay with a 180-day half-life to each assay's contribution to the aggregate, so a clean test from 2024 counts, but not as much as a clean test from last month. This is why "last reviewed" dates matter and why our leaderboard is never frozen.
A vendor with clean results on one peptide has shown us less than a vendor with clean results across ten peptides from three categories. Coverage breadth is a small but real input because a narrow catalog can hide weaknesses in specific synthesis chemistry. We cap the benefit at fifteen unique peptides covered.
When a vendor fails a blind test, what do they do? A thirty-minute response with new batch documentation and an offer to retest beats a hostile legal letter, and beats silence. We hand-score this based on the published remediation note on each failed assay. Small weight because failures are rare enough to keep this column sparsely populated; big signal when it is populated.
A vendor with one perfect test on one peptide does not have the same signal as a vendor with twenty passes across fifteen peptides. To prevent a lucky single test from producing a leaderboard-topping score, we bound the public composite with a Wilson 95% lower confidence interval for vendors with fewer than ten assays on file.
Practically: a vendor with one passing test shows up with a composite that is roughly the score that its single test would produce, penalized by the uncertainty of a sample size of one. As assays accumulate, the confidence bound loosens and the published number climbs toward the point estimate. At n ≥ 10, the confidence penalty is negligible and the point estimate is published as-is.
We show the sample size (N) next to every composite on the leaderboard for this reason. A high score with a small N is a preliminary ranking, explicitly labeled as such. A high score with a large N is the thing you actually want to buy from.
Until our own blind-purchase testing program clears its first few dozen assays, we ground preliminary rankings in publicly published COAs from Finnrick Analytics. Finnrick is currently the closest thing this category has to an independent lab of record. Their public COA catalogue covers many of the same suppliers we track, on many of the same peptides, under a per-test published pass/fail model.
Our debt to them is real and we want to be explicit about it. Where a Finnrick COA appears in our aggregate, it is flagged as third-party-public on the underlying assay record, attributed to Finnrick by name, and linked back to their published report. We do not scrape or republish their catalogue in bulk; we cite individual reports, one at a time, as the evidence base for a specific claim on a specific supplier page.
What our formula adds beyond Finnrick's per-test pass/fail is:
None of that is a knock on Finnrick. It's the same data, re-aggregated to answer a different question: not "did this one sample pass?" but "should I trust this supplier next month?"
◉ This list is live. Each row links to the Open Assay assay record (which documents how we're using the citation) and to the source report on Finnrick.
Janoshik Analytical plays a different role in our data than Finnrick. Janoshik publishes per-test COAs searchable by batch-specific verification keys at verify.janoshik.com— there is no per-vendor composite grade. We cite Janoshik here as attestation: the vendor has opted into a verifiable per-batch testing regime, and a researcher can cross-check any specific lot using the verification key on the vendor's COA. These citations are recorded as inconclusive rather than pass/fail because a single attestation doesn't assert an outcome on a specific test — it records that outcomes are verifiable on demand.
The formula, weights, decay rate, and confidence bounds are fixed in code and reviewed quarterly. Changes to any of them produce a public methodology changelog entry on /methodology, and every affected supplier is notified by email before the change goes live. We consider the ability to show our work, unprompted, to be the brand.