My second doctorate, done at Harbin Medical University with postdoctoral work at Jiangsu University, asked how a small non-coding RNA made by a bacterium could suppress the growth of a lung cancer cell line. To answer that question I built a panel of assays: a specific proliferation assay, a specific set of markers, a specific comparison against a control. The panel answered the question I designed it to answer. It said nothing about apoptosis pathways I did not measure, or the other things that cell line was doing while I watched only one of them. That is not a flaw in the experiment; it is what an experiment is. A well designed panel is a fence around a question, and everything outside the fence stays invisible, not because it was checked and found absent, but because nobody looked.
Why is a silent certificate of analysis a problem?
A certificate of analysis only reports tests that were actually run. A parameter absent from the page usually means nobody specified or paid for that test, not that it was checked and passed. Reading only the results column confirms the questions asked were answered; it confirms nothing about the questions that were never asked at all.
A certificate of analysis is the same kind of document, and I did not see that clearly until I started reading them for a living instead of writing experimental protocols for a living. It reports the result of a defined panel of tests, run against a defined specification, on a defined batch. The panel was set by someone, usually the manufacturer, sometimes against a pharmacopoeial monograph and sometimes against an internal specification that only loosely tracks one. Whoever set it decided, deliberately or by default, which questions this batch would be asked. Everything else about the batch remains exactly as unknown after the certificate as before it.
A certificate is a list of questions, not a verdict
Read plainly, a certificate of analysis looks like a verdict. A column of parameters, a column of results, a column of limits, and every row inside its range. The visual effect is completeness. The actual content is a set of answers to a set of questions that someone, at some point, decided were the ones worth asking of this substance, in this form, from this supplier. Identity, assay or potency, related substances, water content, particle size, microbial limits: which of these appear, and to what depth, depends on the monograph invoked and on what the buyer specified in the purchase agreement, if the buyer specified anything at all.
Two certificates for the same active ingredient, from two manufacturers each claiming to meet the same pharmacopoeial standard, can legitimately differ in which impurities they name individually and which they fold into a single unspecified total. Neither certificate is wrong. They are answering slightly different versions of the same question, and the difference sits exactly where a reader in a hurry will not look: in the list of what is not there.
Three states, and only two of them are visible
A test on any given parameter can end in one of three places: it was run and passed, it was run and failed, or it was never run. The first state fills most of a certificate. The second almost never appears, because a batch that fails a specified test is, in a functioning supply chain, the batch that does not get shipped or does not get a certificate written for it at all. The third state, not run, produces no row, no number, no flag. It is the only one of the three that looks exactly like a clean result to someone scanning a page of passes, and it is by far the most common reason a certificate says less than a buyer assumes it says.
Why the gaps are usually not fraud
It is tempting, once you notice how much a certificate can leave out, to read every gap as concealment. Most of the time that is the wrong reading: it teaches a buyer to distrust documents instead of learning to interrogate them, which solves nothing. A manufacturer's default panel usually reflects the specification most commonly requested of that product line, built around whichever monograph its largest markets recognise. A genotoxic impurity assessment tied to a specific synthetic route, an elemental impurity panel matched to the actual catalysts used, or stability data generated for the climate a shipment is actually travelling through are not always part of that default panel, and are not always missing out of concealment. Often, nobody in the chain of purchase orders ever asked for them, so nobody paid to have them run, and the certificate is silent on exactly the question the buyer needed answered.
This is a purchasing failure before it is a laboratory failure. The specification a buyer writes into a purchase order is the question a manufacturer's laboratory will actually answer. A vague specification, or none, buys a certificate that is genuinely accurate and genuinely uninformative about the risks that matter for the market the product is entering.
The gaps worth asking about, every time
- Elemental impurities matched to the real process. A generic heavy metal check is not the same as a panel built around the catalysts a specific route actually uses.
- Genotoxic impurity assessment tied to the synthetic route. This depends on the chemistry actually used to make the batch, not on a template shared across a manufacturer's product line.
- Stability data for the climate the shipment will pass through. Data generated for one storage zone does not describe a route through a much hotter port and a slower customs process.
- Residual solvents matched to the actual process, not a default list. A synthesis that changed suppliers or steps can change which solvents are relevant without the certificate's template changing with it.
A gap is a question, not an accusation. The honest next step is a specific request back to the supplier for the missing data, not a conclusion drawn from its absence. Most gaps close in one email. The ones that do not close are the ones worth remembering.
Reading for the shape of what is missing
The habit that transferred most directly from a research career is not knowing more chemistry than a buyer without a science background. It is a reading practice: before looking at any result, list what the relevant monograph or specification requires, then check the certificate line by line against that list rather than against its own claims about itself. What is present and in range needs no further attention. What is absent from the list needs a question sent back to the supplier, phrased as a request for the missing data rather than an accusation, because in most cases the honest answer is that nobody asked before.
This is slow, exact work, closer to auditing a methods section than to reading a summary. It is also, in my experience, the single largest source of avoidable risk in the documents that cross a procurement desk, because a reader who checks whether every number sits inside its stated range and stops there has confirmed that the questions asked were answered correctly, and nothing about the questions that were never asked.
This is the specific gap PharmaTrust, the software I built at NJMC, was written to close. It takes a certificate of analysis and the specification it claims to meet, and reports where the two align, where a result sits outside its stated range, and where the specification calls for a parameter the certificate simply does not carry. It analyses documents, not medicines, and does not replace the judgment of a quality team reading the same page with the market and the patient in mind. What it does is make the missing rows visible on the first read instead of the fifth, usually the difference between a question asked before a shipment leaves and one asked, if at all, after something has already gone wrong.
What silence costs in a market with no second check
In a market with an active inspectorate and laboratories that routinely retest imported batches, a gap in a certificate is one layer of a system with several other layers behind it. In a market where the buyer's own review is functionally the last check the batch will ever receive before it reaches a patient, that same gap is the whole system, for that question, for that batch. I have bought for both kinds of market, and the discipline of reading for absence matters in both, but it is not optional in the second kind. There, a silent certificate is not a smaller version of the problem. It is the entire problem, sitting unread in a column that was never filled in because nobody, at any point in the chain, thought to ask.