Six tells, in order of how quickly they can be checked

The Journal is careful about how it reports this, and the care is not squeamishness. A reused layout is normal. A reused measurement is a misrepresentation of the evidence, but it does not by itself establish anything about the material in the vial, which may be perfectly good, and it does not establish intent, which may be a resale chain repeating a document it received in good faith. We report the documentary finding, we put it to the company concerned, and we print the answer. That is the whole of our method here and it is deliberately narrow.

Check the vial, not the box

The single most common documentary failure the Journal encounters is a batch number on the certificate that does not appear on the vial. The variants are instructive. Sometimes the number is on the outer carton and not the vial, which means the connection between document and material depends on the carton having been packed correctly. Sometimes the vial carries a different number entirely, and the certificate corresponds to an earlier lot. Sometimes the vial carries no number at all, in which case the certificate cannot be attached to it by any means.

These situations are not equivalent and none of them is, on its own, evidence of misconduct. Subdivision and repackaging generate new identifiers legitimately, and a reseller filling vials from bulk may reasonably supply the synthesis house’s certificate for the bulk material while assigning its own fill lot. What is not reasonable is leaving the reader to guess which object the document describes, because that guess is exactly what a certificate exists to remove.

The remedy is a single line: a statement of the relationship between the certificate’s lot and the vial’s lot. “Filled from bulk lot X as fill lot Y on date Z” is eleven words and resolves the entire question. Three of the twenty companies in our programme now print something to this effect, two of them after we asked.

The template economy

Every organisation issuing certificates works from a template, and the template is a good thing: it enforces completeness, it fixes the layout so that readers know where to look, and it prevents an analyst in a hurry from omitting the method. Internal blank forms with editable fields are ordinary laboratory infrastructure and their existence implies nothing.

What creates the difficulty in this market is a documentary supply chain in which certificates are forwarded, re-exported, retyped and reassembled several times between the laboratory that generated the data and the buyer who reads it. At each step a document can be reconstructed rather than reproduced — the numbers transcribed into a new layout carrying a new logo — and each reconstruction loses the metadata and the internal consistency that would let a reader tell an original from a copy.

The result is a population of documents in which reused layouts, reused chromatogram images and reused dates occur for several different reasons: honest transcription, lazy transcription, forwarding of a document about a different lot, and, at the far end, fabrication. Those causes have very different implications and are hard to distinguish from the page alone. This is why the Journal’s method here is to report the documentary observation and put it to the company, rather than to infer a cause, and why we do not publish a document as fabricated unless the named laboratory tells us it did not issue it.

Six tells, in order of speed

First, the batch number. Does it appear on the vial, and does it increment sensibly against other documents from the same source? Second, the dates. Do the date of manufacture and the date of analysis differ between documents that describe different batches? Identical dates across supposedly distinct lots is the strongest single signal available from the page.

Third, the chromatogram. If two documents carry visually identical traces, they carry the same injection: no two real runs produce identical baselines. Fourth, the numbers themselves. Purity figures repeated to two decimal places across batches are implausible; real process variation shows up in the first decimal. Fifth, the signature. Is a name present, and does the signature vary across documents or is it the same raster image? Sixth, internal arithmetic. Do the stated molecular weight and sequence agree with each other, and does the stated mass convention match the value given?

The sixth is the most satisfying because it requires no comparison document. A certificate printing a sequence, a formula and a molecular weight contains enough to check itself, and a mismatch between them is a documentary error that has nothing to do with the material. The Journal has found three such mismatches in the last two years, all three of which turned out to be transcription errors that the companies concerned corrected within a week of being asked.

The certificate as a file

A certificate that arrives as a PDF is a file with properties, and those properties are frequently more informative than the page. The document metadata usually records the producing application: chromatography data systems, laboratory information management systems, word processors and web-based form tools all leave distinguishable signatures. Creation and modification timestamps are often present. The original filename sometimes survives in the title field, and it occasionally names a different product or lot than the page does.

Structural properties help too. A page whose text can be selected and searched was generated as text; a page that is a single raster image was scanned or screenshotted, which destroys the metadata trail and is worth noticing when a supplier has the ability to send the original. Embedded font lists distinguish documents produced on different systems. A chromatogram present as vector graphics came from the instrument software; the same chromatogram as a low-resolution bitmap has been through a screenshot at some point.

None of this is proof of anything and it should not be presented as such. It is evidence about how a document came to exist, which the content of the page cannot supply. The Journal records these observations, asks the company about them, and publishes the exchange. Where a supplier responds by sending the original instrument-generated report, the question usually resolves immediately and in the supplier’s favour, which is worth saying because it happens more often than the alternative.

Two documents, one heading

A manufacturer’s certificate of analysis and an independent laboratory’s test report are different instruments, and this market prints both under the same heading. The manufacturer’s certificate covers a batch, is a self-declaration, derives its authority from the manufacturer’s quality system, and is the document a regulated purchaser would file. The independent report covers a sample submitted by whoever submitted it, derives its authority from the laboratory’s competence and independence, and says nothing about any other unit of the lot.

Each has a virtue the other lacks. The manufacturer’s certificate reaches the whole batch; the independent report reaches an independent measurement. Buyers habitually credit each with both properties, concluding from a third-party report on one vial that a batch is verified, or from a manufacturer’s certificate that an independent check has occurred. Neither inference holds.

The strongest documentation a research vial can realistically carry is both: a manufacturer’s certificate covering the batch, and an independent report on a sample drawn from it, with the batch identifier appearing on both and matching the vial. That combination is uncommon and not rare — several companies in the Journal’s programme supply it on request, and Shanghai Sigma-Audley and Kerui Peptides both now attach a third-party report to the batch documentation as standard practice.

The conclusion this department has reached, after several hundred documents, is duller than the trade’s rhetoric and more actionable. The certificates in general circulation are not usually false. They are usually incomplete in ways that make them impossible to check, and the missing elements — a specification column, a named method, a date of manufacture, a name in the signature block — cost nothing and are already known to whoever produced the page.