How ResearchVerify scores a COA

1. Why this rubric looks the way it does

The methodology is the product. Everything else is plumbing.

A scoring system for Certificates of Analysis only matters if buyers and vendors both believe it reflects how the market actually behaves. ResearchVerify is built on two principles that shape every other decision on this page.

The rest of this page documents how those principles show up in the math.

2. The composite score

A weighted average of four independent measurements, on a 0–10 scale.

Testing protocol carries the most weight because it's the part vendors most often shortcut. Chain of custody and the Conformance Index are smaller weights because they're correctness checks, not capability measurements. A perfect score in all four components is 10.0; a typical good record sits in the 7.0–8.5 range.

Most records score below that range, and the database-wide average sits near 6.0. That is a fact about what gets submitted, not a judgment on the compounds. Because 40% of the score rewards testing completeness, a certificate reporting only purity on a single vial — no mass-spec confirmation, no net-peptide content, no sterility or endotoxin panel, one vial instead of several — is capped well below 10 even at 100.00% purity. Single-vial, purity-only certificates are the bulk of what researchers upload, so the corpus average reflects thin testing far more than it reflects bad product. A high score means a vendor tested thoroughly and the numbers were clean; a middling score usually means the testing was minimal, not that the peptide was poor. The score is only as complete as the certificate behind it — read the two together.

3. Testing protocol (40%)

Purity tier plus the depth of analytical tests performed.

The compound class system

Different compounds have different achievable purity ceilings because of how they're manufactured. A small synthetic peptide like BPC-157 routinely reaches ≥99.5% purity; recombinant HGH (191 amino acids, prone to misfolding and dimerization) reaches 94–96% in pharmaceutical-grade material. Scoring both against the same scale would over-reward small peptides and under-reward large proteins. ResearchVerify uses three compound classes.

Class A: Small peptides (≤50 amino acids)

Examples: BPC-157, TB-500, GHK-Cu, Semaglutide, Tirzepatide, Retatrutide, Ipamorelin, CJC-1295, MOTS-c

Class B: Large proteins (>100 amino acids)

Examples: HGH / Somatropin (191 aa), IGF-1, MGF

Example: HGH measured at 94.7% scores Tier 2 (3.5 pts). This is normal for recombinant HGH; scoring it as "poor" would penalize an achievable result for the compound class.

Class C: Lipidated peptides

Examples: Semaglutide (when lipid chain is separately resolved), Liraglutide

Additional tests beyond purity

Purity alone is not enough. A COA that shows only HPLC purity is incomplete relative to one that also confirms molecular identity, peptide content, and microbial safety. Each additional test adds points.

Maximum is 11 raw points (4 purity + 7 additional), normalized to a 10-point scale before the 40% weight applies.

4. The overfill rule, explained

This is the methodological choice people ask about most. Here's the full reasoning.

If a vial labeled 10 mg actually contains 11.4 mg of peptide, what's the right way to score that?

Some verification systems treat any divergence from the label as a problem — over or under. The reasoning sounds rigorous: "the label is the spec; missing the spec in either direction is sloppy manufacturing." On its face, that's defensible.

But it doesn't reflect how the market actually behaves, and on closer inspection it doesn't reflect how peptide manufacturing actually works either. Two reasons.

Reason 1: Lyophilized peptides include non-peptide mass

When peptide is freeze-dried into a vial, what ends up in the vial is the active peptide plus trifluoroacetic acid (TFA) counterions, residual moisture, and trace salts. The total dry mass weighed into the vial is larger than the active peptide content. A responsible manufacturer accounts for this: they weigh in more total powder so that the active peptide hits the labeled amount. From the manufacturer's process, that looks like overfill. From the buyer's perspective, that's the labeled amount being honest.

Reason 2: Buyers price by label, not by precise active content

If a buyer pays for 10 mg labeled and receives 11.4 mg measured, the buyer has received more than they paid for. There is no scenario in which this is a defect. Penalizing the manufacturer for it forces them to dial overfill down, which transfers value from buyer to vendor, which is the opposite of what a buyer-protection rubric should do.

The scoring table

Edge cases the rubric handles

Class B unit conversions: HGH and similar large proteins are labeled in International Units (IU), not milligrams. The rubric converts mg to IU using accepted ratios before computing variance. For HGH: 1 mg ≈ 3 IU. A vial labeled 36 IU measured at 13 mg of HGH converts to 39 IU actual, scoring as +8.3% overfill.

Implausible overfill (>50%): Extremely large overfill values (e.g., a vial labeled 5 mg measured at 15 mg) trigger a separate Conformance Index flag (DI-05). The label points are still awarded, but the record is flagged for human review because such large divergences usually indicate a label or measurement error, not generous overfill.

Large overfill as a dosing note (not a penalty): The reasoning above assumes the ordinary case, where a modest measured surplus is counterions, moisture, and salt rather than extra drug — no dosing consequence, and no flag. Large overfill is different: a vial measured well above its label, past what excipient mass alone explains. We still don't score it as a defect — receiving more than you paid for is not a shortfall, and that stance doesn't change. But it is worth surfacing, because if the surplus is active peptide, anyone dosing by drawing a fraction of the vial rather than reconstituting to the labeled amount will pull a proportionally larger dose than intended, and on a potent compound that is not trivial. Large overfill is therefore carried in the Conformance Index (flags QV-02 / QV-02b, and DI-05 above 50%) and shown on the record as a dosing note — flagged for the reader, never subtracted from the score. The safe response either way: reconstitute to the labeled amount and dose from that.

5. Label alignment (30%)

How well the label corresponds to what was measured.

Maximum 10 points, applied at 30% weight in the composite.

6. Chain of custody (20%)

Whether the sample's path from manufacturer to lab is documented.

This component is about whether a buyer can audit what happened to the sample before testing. Multi-vial submissions, dated receipts, and tamper-evident seals all increase confidence that the test result actually reflects what's in the buyer's vial — not a hand-picked best sample.

Raw maximum is 115 points (the elements can stack), normalized to 10 before the 20% weight applies.

7. Conformance Index (10%)

Data anomalies that suggest the COA is wrong, manipulated, or inconsistent.

The Conformance Index is a 0–100 score built by deducting points for specific anomalies. Each anomaly has a documented code so flagged records can be audited and disputed. We publish every code we use.

Data integrity anomalies

Temporal anomalies

Note: a 1-day turnaround (received one day, tested the next) is normal lab workflow and carries no penalty. T-01 flags same-day-only because that's the pattern associated with fabricated COAs.

Statistical and documentation anomalies

How the CI score maps to the composite

8. BS detection

A defensive tripwire that scans every COA for pseudoscientific claims. In 10,044 records to date, zero have triggered.

Some COAs contain claims that aren't measurable scientific properties — "quantum purity," "vibrational frequency," "energy alignment." These claims are not analytical results; they are marketing language inserted into a document that should contain only analytical results. Records containing them would receive zero credit for the disputed metric and a public flag.

The validator has scanned every record in the database against the categories below. To date, no submission has triggered a BS flag. The detection logic remains active as a tripwire — both because submission patterns change over time, and because the absence of triggers is itself useful information about the labs the community is sourcing from.

Labs that accumulate two or more BS detections across separate records receive a "frequent invalid claims" badge attached to their lab profile. The badge is removable when subsequent records show no further pseudoscientific claims.

9. Credibility modifiers

The final composite score is multiplied by a modifier based on testing source.

The modifier is applied after the composite is calculated, so it scales the entire score rather than any one component.

10. Worked example, end to end

A real Janoshik COA, scored from raw measurements to final composite.

Sample inputs

• Compound: TB-500 (Thymosin Beta-4) — Class A small peptide
• HPLC purity: 99.085%
• Label: 10 mg; measured content: 11.38 mg (+13.8% overfill)
• Batch ID present, single vial tested, receiving date documented
• No mass-spec, NPC, sterility, or endotoxin testing on this COA
• No tamper seal, no lab custody seal, no vial image
• Third-party lab (Janoshik), not ISO 17025 accredited

Component calculations

# Testing protocol (40% weight)
Purity 99.085% (Class A, Tier 2)        = 3.5 pts
Additional tests: none                  = 0.0 pts
Raw total: 3.5 / 11 max → normalized    = 3.18 / 10

# Label alignment (30% weight)
Product name matches                    = 3 pts
Quantity: +13.8% overfill (full credit) = 4 pts
Batch ID (20260428) present             = 2 pts
Expiration date absent                  = 0 pts
Raw total: 9 / 10                       = 9.00 / 10

# Chain of custody (20% weight)
Batch ID on vial                        = 15 pts
Single vial tested                      = 15 pts
Receiving date documented (04 MAY '26)  = 15 pts
No tamper seal, no custody seal, no img = 0 pts
Raw total: 45 / 115 → normalized        = 3.91 / 10

# Conformance Index (10% weight)
D-01: vendor field is an email address  = −5 pts
FORENSIC-02: moderate manipulation flag = −5 pts
Base 100 − 10 = 90 (Good)               = 9.00 / 10

# Weighted composite
(3.18 × 0.40) + (9.00 × 0.30) + (3.91 × 0.20) + (9.00 × 0.10)
= 1.27 + 2.70 + 0.78 + 0.90
= 5.65 / 10

# Credibility modifier: third-party lab (×1.0)
Final RV Score = 5.65 / 10

This record scores in the mid range — solid purity, good label discipline (the overfill helps), but limited testing depth and minimal custody documentation. If the same vendor had submitted multi-vial testing with mass-spec confirmation and an endotoxin screen, the score would lift into the 7s. That's exactly the signal a buyer would want to see when comparing two vendors of the same compound.

11. What this rubric does NOT yet do

Honest acknowledgment of where the methodology is incomplete.

A scoring rubric is only as credible as its acknowledged limitations. ResearchVerify is open about what it currently is and isn't.

If you find a record where the rubric appears to score something wrong, the right action is to submit a community note on that record. Disputed records are reviewed; the rubric itself updates when systematic gaps are found.

12. Version history

The methodology has been iterated openly. Every change is documented.

Methodology questions, dispute submissions, and proposals for rubric changes can be sent through the community channels linked in the main site footer.