Evidence-First Scoring Is Live: One Formula, Every Term Published

PeptideBenchmark's vendor scores now put verifiable lab evidence at the center: one evidence scale for all 400+ vendors, outside review sources bounded to a small adjustment, per-vendor term receipts on every profile, and machine-verified reproducibility on every deploy.

---

Every vendor score on PeptideBenchmark is now evidence-first: one formula for all 400+ vendors, with verifiable lab evidence as the backbone and every term published — on the methodology page globally, and on each vendor profile individually, where the Rank Composition receipt sums to the displayed score.

What we fixed

The previous formula did its job when this site was young, but it had two flaws we no longer wanted to live with — and we’d rather name them ourselves than have someone else do it:

It leaned on outside benchmarks as its backbone. A vendor’s standing depended heavily on whether third-party listing sites covered them, and how. That made scores fragile in ways that had nothing to do with what’s in a vial: a vendor absent from those sites was stuck near a neutral anchor no matter how strong their verifiable testing was.

Thin data blended toward neutral. With little information, the old math pulled scores toward ~7 — which sounds harmless until you notice it works in both directions. A vendor with a genuinely poor outside rating on limited volume could still display a mid-7 score. The blend was hiding adverse information, and a review site should never do that.

What the score is now

score = verified-evidence backbone
      + bounded review adjustment (max +0.5 / −1.5)
      + curated bonus − overseas adjustment

The backbone comes from what we can verify ourselves: certificates that re-check at the issuing laboratory, verified purity above 98%, and coverage across the catalog — pricing plays no role in scoring — with a bridge for independent lab-test data on vendors we haven’t fully reviewed, and a floor system so displayed trust bands and scores can never contradict each other. The exact weights, caps, and floors are published in full.

Outside review sources — lab-test aggregators, review directories, Trustpilot, Google — still appear on profiles with their own numbers. But their effect on the score is now a single bounded adjustment: at most +0.5 combined, up to −1.5 downward, scaled by review volume. Reviews can warn you about a vendor. They can no longer buy one a ranking.

And where we can’t verify enough to score honestly, we don’t score: fourteen vendors are shown unranked instead of carrying an invented number (five of them scored under the old formula; their profiles still show that history). They regain a score the moment verifiable evidence exists.

How the cutover was done

We didn’t flip a switch. The new formula ran for weeks alongside the old one, computing both scores for every vendor, gated on continuity checks: median absolute change of 0.34 points on vendors not being deliberately re-ranked, rank correlation 0.96, and zero unexplained rating-band migrations. All but one of the larger score moves carry a published, data-backed reason (usually: the score had been riding an outside listing, or the old blend was hiding adverse data); the one residual move we reviewed and accepted by hand before flipping anything. The formula then ran as a labeled public preview on every profile before becoming the live score.

Vendors whose score moved show their previous-formula score alongside the new one during the migration period, so nothing changes silently. And the same guarantee that has protected this site all along still holds, now for the new formula: before any deploy goes live, an automated check recomputes every score from the same published data files and refuses the release on any mismatch.

What this means for readers

The top of the board now belongs to vendors with deep, re-checkable evidence — certificates you can verify at the laboratory yourself (here’s how) — not to whoever accumulated the most listings. When you see a score, the receipt is one click away. When you see a vendor without one, that’s us being honest about what we don’t know yet.

Questions about a specific vendor’s move? Every profile’s Rank Composition panel has the term-by-term answer, and the methodology page has the formula behind it.

Common questions

What changed in PeptideBenchmark's vendor scores?

The score's backbone changed from outside benchmark ratings to PeptideBenchmark's own verified testing evidence — certificates that re-check at the issuing laboratory, verified purity, and catalog coverage; pricing plays no role in scoring. All outside review sources together are now a bounded adjustment of at most +0.5 upward and −1.5 downward. Every profile shows a term-by-term receipt that sums to the displayed score.

Why did some vendors' scores drop?

Mostly for one of two published reasons: the old formula blended thin data toward a neutral anchor (which could hold a vendor near 7 on very little information, or soften a genuinely poor rating), or a score was riding an outside listing rather than verifiable evidence. Every affected profile shows its previous-formula score alongside the new one during the migration period, with the receipt explaining the difference.

Can reviews still improve a vendor's score?

Only slightly. All review sources combined can add at most +0.5 points, scaled by review volume — but they can subtract up to −1.5. Reviews can warn; they cannot buy points. Verifiable lab evidence is the only way to a top score.

How do I know the scores are computed as described?

Two ways: every vendor profile publishes its own term receipt under Rank Composition, and an automated check recomputes all 400+ scores from the same published data files before any deploy goes live, refusing the release on any mismatch. The full formula, with every weight and cap, is on the methodology page.

What happens to vendors with no verifiable evidence?

They are shown unranked rather than assigned a number. Their review signals stay visible, and they gain a score as soon as verifiable testing evidence exists. Fourteen vendors are unranked under the new formula; five of them carried a score before the change, which their profiles still disclose.