Methodology
Every number on this site, and how it was produced. Where a figure cannot be reproduced from this page plus the published dataset, that is a defect and it goes in the corrections log.
1. What becomes a record
A submission becomes a record when it carries: a named laboratory, a lot number, a compound, at least one quantitative result, and a documented interval between purchase and assay. Submissions failing any of those are refused. The largest single category of refusal remains the screenshot — a cropped or photographed chromatogram with no lot number and no letterhead is not a record and the index does not hold it.
2. Normalisation
Results arrive in whatever format the laboratory uses and are mapped into one schema of 39 fields. Two normalisations are applied and both are visible in the data:
- Retention times are normalised to the index reference method RG-2 (30-minute gradient) so that a retention window means the same thing across laboratories. The transform is linear in gradient duration and is documented on its own page. The record shows both the observed and the normalised value.
- Nothing else is adjusted. In particular purity results are not bias-corrected. Laboratory offsets are published and can be applied by a reader; the index does not silently apply them, because a corrected number that looks like a measurement is worse than an uncorrected one beside its offset.
3. The numeric coherence contract
Every record satisfies these, and the build asserts them row by row. A record that breaks one is a defect in this site, not a finding about a supplier.
| Invariant | Rule |
|---|---|
| Chromatographic closure | purity + Σ(impurity area %) = 100.00 ± 0.02 |
| Impurity ordering | largest single ≤ total impurities, and is row one |
| Peak table ↔ figure | The chromatogram is drawn from the peak table. There is no independent draw. |
| Gravimetric mass balance | water + counter-ion + residual + peptide content = 100.00 ± 0.02 |
| Target-compound content | target = total peptide content × purity ÷ 100 |
| Vial recovery | recovered = gross fill × target ÷ 100 ; % of claim = recovered ÷ claim × 100 |
| Mass accuracy | observed = theoretical × (1 + ppm ÷ 1e6) |
| Charge states | m/z = (M + z × 1.007276) ÷ z |
The consequence worth stating: anomalies are never manufactured by corrupting the index's own arithmetic. Every flag is a comparison between a supplier-declared figure and a measurement, between a measurement and a published threshold, or between a distribution and its own shape. The index's numbers always add up; the disagreements are between the index and somebody else.
4. Statistical definitions
- Quantiles are type 7 — linear interpolation between order statistics, the R default.
- Medians, not means, throughout. The purity distribution is compressed at the top with a long left tail; a mean over that shape is dragged by exactly the observations you most want to see individually.
- Spread is the interquartile range, or for small sets the max−min range, always labelled as which.
- Robust scale where used is the median absolute deviation, scaled by 1.4826 to be a consistent estimator of σ for normal data.
- Box plots are Tukey: box Q1–Q3, whiskers to the extreme observations inside 1.5 IQR of the box, and every point outside the fences plotted individually rather than summarised.
- Paired analysis for inter-laboratory work is Bland–Altman: mean difference, 95 % confidence interval on that mean, and 95 % limits of agreement at ±1.96 sd.
- Histograms use a fixed 92–100 % domain and a 0.25 pp bin width everywhere on the site, so any two can be compared by eye.
5. The n= convention
The n= badge next to a statistic is the number of records it was computed from. Below n=8 the badge is amber. Below n=3 the statistic is suppressed and the individual values are shown instead. This is enforced by the rendering layer rather than left to the author, so there is no page on this site where it can have been forgotten.
6. Verification status
A record's status describes what its measurements can support, and nothing else.
| Status | Means |
|---|---|
| VERIFIED | Purity, identity at high mass resolution, and at least one orthogonal measurement (content, water or counter-ion) were all determined, on a method meeting the index specification for the compound class. The record is complete as far as the index defines completeness. |
| IDENTITY CONSISTENT | Purity and identity were both determined, but identity was confirmed at unit mass resolution. The observed mass is consistent with the named compound and does not exclude near-isobaric alternatives, which are named below. |
| PURITY ONLY | A purity figure was determined and no identity confirmation was ordered. The record describes the chromatographic homogeneity of the material submitted and does not establish what that material is. |
| METHOD LIMITED | The method used does not meet the index specification for this compound class — most often a gradient shorter than the class minimum, or reverse-phase chromatography applied to a substance for which it is not the reference technique. The result is published with the limitation named and is excluded from the tightest aggregates. |
| PARTIAL | A measurement the compound requires is missing — a metal determination on a metal complex, a PEGylation distribution on a conjugate, an anomeric ratio on NAD+. The record is published because what was measured is still informative, and marked partial because what was not measured is essential. |
| N/A | The substance is not a single defined chemical entity, so purity, monoisotopic mass and retention time are not defined quantities for it. The record reports what can honestly be measured and states the rest as undetermined. |
7. Provenance weighting
The sample was purchased through the supplier's ordinary retail channel by the submitter, at the submitter's own cost, and forwarded to the laboratory without the supplier's involvement or knowledge. The supplier had no opportunity to select the vial. This is the provenance class the index weights most heavily and the only one used in the sensitivity analyses on supplier scorecards.
The sample was provided by the supplier, either as a free sample or as material sent directly to the laboratory at the supplier's request. The supplier chose which vial was tested. Across the 88 lots in this index where both provenance classes are available for the same lot, supplier-supplied samples read a mean 0.31 pp higher than independently purchased vials (95 % CI 0.19 to 0.43). That difference is small on average and occasionally large; rule AX-R10 flags the individual cases. The result below is published in full, and it should not be treated as equivalent to a buyer-sourced result.
8. What the index refuses to compute
- A composite supplier score. See the scorecard rubric for why.
- A market-share estimate. Submission volume is a property of who tests, not of who sells.
- A trend from fewer than four quarters of a compound-supplier cell. Three points and a ruler is not a trend.
- Anything about a compound the index holds no records for. 3 catalogued compounds currently have none, and their pages say so instead of extrapolating from the class.
9. Reproducibility of the site itself
The whole site is generated by one script from a string seed. Two runs produce byte-identical output; there is no Math.random() anywhere in the generator. Every figure, every chromatogram and every baseline noise trace is derived from the record's own identifier, so a record's figure is stable across rebuilds and a change in a figure means a change in a number.