How to Read a Certificate of Analysis, and the Nine Places a Number Can Mislead You

A certificate of analysis is the most requested document in materials purchasing and among the least well read. It is treated as a pass-or-fail stamp when it is really a dated record of specific measurements performed on a specific quantity of material by a specific method. Every one of those four qualifiers is load-bearing, and a reader who does not know which one is doing the work in a given line can be handed an accurate document and still reach a wrong conclusion. This is a working guide to reading one properly: what identity testing settles, what purity testing cannot, why the headline percentage and the actual content of the container are two different numbers, how a date on the page is earned, and the recurring ways a technically truthful certificate misleads a careful person.

A certificate describes a lot, not a product

This is the distinction almost everybody collapses, and nearly every other misunderstanding on this page follows from it. A certificate is issued against a lot, sometimes called a batch: one discrete quantity of material produced in one run under one set of conditions. It is not issued against a product line, a catalogue entry, a supplier, or a brand.

Two lots of the same nominal material, from the same facility, made six weeks apart, are two different objects analytically. They may differ in residual solvent, in water content, in counterion ratio, and in the profile of related substances present at low level. Those differences arise from ordinary variation in a real process: a slightly different coupling efficiency, a longer hold before drying, a new drum of starting material. A certificate for one lot says nothing whatsoever about the other, and a supplier who shows you a certificate for a lot they are not shipping has shown you a document about material you will never receive.

So the first question about any certificate is not whether it looks good. It is whether the lot number on the document matches the lot number physically on the container in front of you. Not the product code. Not the item number. The lot. If those two strings differ, nothing else on the page is worth reading, because the document and the material are unrelated.

There is a softer version of the same failure that is much more common than outright substitution: the representative certificate. This is a real document, from a real analysis, of a real lot, offered as an illustration of what the material is generally like. Sometimes it is even labelled as such in small type. It is not evidence about your container, and the moment it is filed as though it were, the paper trail has a hole in it that will never be visible again.

The three separate questions, and why they get collapsed

Analytical documentation answers three questions that feel like one question and are not:

• Identity. Is the substance present the substance named?
• Purity. Of the material detected, what proportion is the named substance rather than something related to it?
• Content. Of the mass in the container, how much is the named substance at all?

Most readers treat a single high percentage as the answer to all three. It is usually the answer to the second, sometimes the answer to the third, and almost never the answer to the first. The sections that follow take them in turn, because the gaps between them are exactly where an accurate document produces a wrong belief.

Identity: what a mass measurement does and does not settle

Identity is normally established by measuring molecular mass. The instrument ionises the sample, sorts the ions by mass-to-charge ratio, and reports what it finds. If the observed mass matches the calculated mass for the expected molecule, that is strong evidence the molecule is what the label says.

Strong, but not conclusive, and the reason is worth understanding. Mass is not structure. Two different arrangements of the same atoms have the same mass. For a chain molecule this is not a theoretical worry: reverse a sequence, swap two residues, substitute one component for another of identical mass, and the number the instrument reports does not move. A mass measurement alone cannot distinguish any of those from the intended material.

What closes the gap is a technique that interrogates structure rather than total mass. Fragmenting the molecule and reading the masses of the pieces constrains the order of the components, not merely their sum. Sequence-level confirmation is a different and more expensive test than a single mass reading, and a certificate that carries one is telling you something a certificate that carries the other is not.

In practice, a mass figure consistent with expectation plus a chromatographic profile consistent with a single dominant species is adequate evidence for most purposes. The thing to notice is when a document presents a mass number alone and invites you to read it as proof of identity. It is evidence of identity. It is not proof, and the difference matters precisely in the cases where something has gone wrong.

Purity is a method with a number attached

A figure such as 99.1% is close to meaningless in isolation, because purity is always purity as measured by something. The technique determines what can be seen at all, and everything the technique cannot see is invisible in the number. This is not a minor caveat. It is the whole of what the figure means.

Read that table once more as a list of blind spots rather than a list of capabilities. A certificate reporting a single high number from a single technique has told you that one technique found little of what that technique is able to find. That is a genuine and useful result. It is not a statement about everything in the container.

The wavelength problem

"Purity by HPLC" with no wavelength stated is a common and quietly meaningless construction. Detection at 214 nanometres responds largely to the backbone bonds that every part of a chain molecule contains. Detection at 280 nanometres responds mainly to a few specific aromatic components, and if the molecule contains none of them, the detector is nearly blind to it.

The consequence is direct: the same vial, analysed on the same instrument on the same afternoon, will return different purity figures at those two wavelengths, and both are correct. An impurity that absorbs strongly at the chosen wavelength is over-represented in the result. One that barely absorbs there is under-represented, or simply absent from the number. A certificate that does not state which wavelength was used has not told you what its own number means.

What area percent is, and what it is not

Nearly all reported chromatographic purity is area percent: the area of the main peak divided by the total area of all peaks the detector saw, expressed as a percentage. This is a perfectly respectable measurement, and it contains an assumption that is almost never stated, namely that every component responds to the detector in proportion to how much of it is present.

That assumption is false in general. Different substances have different response factors, sometimes by a large multiple. A related substance that absorbs half as strongly as the main component will contribute half the area it should, and the main peak's share will be flattered accordingly. Correcting for this requires a reference standard for each impurity, which for most materials does not exist, so the uncorrected figure is reported and everyone proceeds. That is defensible. It is only a problem when area percent is read as weight percent, which it is not.

Why the chromatogram itself is the document worth asking for

Everything in the figure above is lost when a result is reduced to one line in a table. Where the integration baseline was drawn, whether the solvent front was included, whether anything was eluting after the run was stopped, whether the main peak is symmetrical or carries a shoulder suggesting an unresolved second component: all of it lives in the trace and none of it lives in the number.

A shoulder is the most instructive example. Two components with similar retention can merge into a single peak that integrates as one. The reported purity is then high and wrong, not because anybody lied, but because the method did not resolve what was there. The trace shows the asymmetry. The table cannot.

The impurity profile: the list matters more than the total

If a material is reported as 99.0% pure, the remaining one percent is not one thing. It is a distribution, and the shape of that distribution carries more information than the total does. Two materials can both be reported at 99.0% and be entirely different objects:

• One has a single related substance at 1.0%. That is a specific, repeatable process outcome. Something identifiable happened at one step, every time, and it can be traced and addressed.
• The other has ten related substances at 0.1% each. That is a general lack of control, or a material that has begun to change in storage, and it is a harder problem because there is no single cause to find.

Which of those is preferable depends entirely on what the single large component is, and that is the point: the total alone cannot distinguish them, so the total alone cannot support the comparison people most often want to make with it.

Specified and unspecified

Documentation distinguishes between specified related substances, which are known, named, individually limited and usually understood to arise from a particular step, and unspecified ones, which are whatever else appeared, held to a blanket limit each. A specification that lists its specified impurities by name, with individual limits, is describing a process somebody understands. One that reports only a single total is describing a measurement, not a process.

Convention also recognises three ascending thresholds, and they are worth knowing because they explain why small peaks are treated so differently from one another. Below the reporting threshold, a peak is not written down at all. Above it, the peak is reported as a number. Above the higher identification threshold, it is no longer acceptable merely to report it; it has to be identified. Above the qualification threshold, identification is not enough either, and the substance has to be justified in its own right. The thresholds are the reason a 0.05% peak and a 0.5% peak attract completely different amounts of work, and the reason a document listing several unidentified peaks above its own identification threshold is incomplete on its own terms.

What the related substances usually are

For chain molecules the common classes are predictable, and recognising them turns a list of retention times into an account of what happened during manufacture and storage.

Truncated and deletion variants are chains missing one component, produced when a coupling step did not go to completion. These tend to sit close to the main peak, because a molecule missing one unit out of many behaves almost identically in a separation, and close-eluting variants are exactly the ones a short or poorly resolved method merges into the main peak.

Oxidation and deamidation products arise after manufacture, from exposure to air and water respectively. Their presence is therefore a storage and handling signal rather than a synthesis signal, which makes them the most useful class to watch across the life of a lot: they are the components whose share should be expected to grow if the material has had a worse history than its documentation assumes.

Protecting-group and reagent residues are leftovers from processing rather than relatives of the molecule. They usually separate cleanly, which makes them easy to see and easy to quantify.

The practical upshot: ask what the largest related substance is, not merely what the total is. A supplier who can tell you which process step produces their principal related substance understands their own material. One who can only quote a total has given you a number from a report.

The mass balance problem: purity is not content

This is the section most readers have never had explained, and it is the one that changes how a certificate is read permanently.

Chromatographic purity is a statement about the organic material the detector saw. It is a ratio between the main component and its relatives. It is not a statement about the total mass in the container, because several things in that container are invisible to the detector entirely:

• Water. Many dried materials are hygroscopic and will take up moisture from the air during handling. Water contributes mass and contributes nothing to a chromatogram.
• Counterion. Materials isolated from a purification step commonly arrive as a salt. The accompanying acid or base is part of the solid, is part of the mass, and does not appear as a peak.
• Residual solvent. Traces of processing solvent survive drying. Measured by a different technique, absent from the purity figure.
• Inorganic residue. Buffer salts carried through from the final step. Same story.

So a material can be entirely honestly described as high purity by chromatography while a meaningful fraction of what is in the container is none of the named substance. The two numbers are not in conflict. They are answers to different questions, and the certificate is only misleading if it offers one while the reader believes they are getting the other.

The date is a forecast, and it is conditional

Every dated certificate carries an implicit prediction: that the material will still meet its specification at some later point. That prediction is earned from stability data, and it is earned under stated conditions. The conditions are not decoration. They are the entire basis of the claim, and a date quoted without them has been detached from its own evidence.

How the date is actually produced

Two study designs sit behind almost every date you will see, and they are not equally strong.

Long-term studies store the material at its intended condition and test it at intervals across months or years. The result is direct evidence. Its disadvantage is that it takes exactly as long as the claim it supports, so a three-year claim requires three years.

Accelerated studies store the material deliberately warm, measure the faster change, and extrapolate backwards to the real storage temperature. This produces a number in weeks rather than years, and it rests on an assumption: that the mechanism of change at the elevated temperature is the same mechanism, only faster, as at the intended temperature.

That assumption fails more often than the convenience of the method encourages people to admit. If warming introduces a different pathway of change, one that is negligible at the real storage temperature, the extrapolation is being driven by a process that will never occur in practice, and the prediction is pessimistic. If warming takes the material through a physical transition, a glass transition or a collapse of dried structure, the extrapolation is not merely wrong in degree; it is describing a different physical state. Accelerated data is useful supporting evidence and a reasonable basis for a provisional date. It is weaker evidence than somebody having actually waited.

Excursions, and why the average temperature is the wrong question

Real storage is not isothermal. Doors open, shipments sit on loading docks, freezers defrost on a cycle. The useful way to think about accumulated exposure is not the arithmetic mean temperature, because change does not scale linearly with temperature. Rates typically rise steeply, so a short warm excursion contributes far more to total degradation than an equal period of equivalent cooling subtracts.

The quantity that captures this properly is the mean kinetic temperature: a single temperature that would produce the same total change as the actual fluctuating history. It is always higher than the simple average, and the gap widens the more variable the storage has been. A material whose average storage temperature looks reassuring can have a mean kinetic temperature well above its stated condition, and the date on its certificate is then describing a history the material did not have.

Freeze-thaw, and the cost of looking

One failure deserves naming because it is caused by diligence. Each time a container is removed from cold storage to be inspected, weighed or sampled, it warms, the headspace reaches equilibrium with room air, and on return it cools again. Moisture condenses on the coldest available surface, which is the material. Repeated conscientious checking can degrade a material faster than leaving it alone, and the resulting change is invisible until someone measures it.

The container is part of the result

A specification describes material in a defined container-closure system, and the system is not incidental to the claim. Three mechanisms do most of the work.

Headspace. The air above the material contains oxygen and water vapour. A container with a large headspace relative to its contents holds proportionally more of both, which is why a part-used container is in a worse position than its original specification assumed, and why the same material split into smaller sealed units is in a better one.

Closure integrity. An elastomeric closure is a consumable. Each puncture is permanent, and the barrier is cumulative rather than binary. Seal integrity is also temperature dependent, because the closure and the container wall do not contract at the same rate.

Light. Where a specification says protect from light, the statement is part of the condition under which every other number was earned. A material transferred from an opaque container to a clear one is no longer stored as its own documentation assumes.

The practical consequence: repackaging invalidates the stability portion of a certificate even when the material is untouched and the handling was careful. The document describes material in a particular system, and that system has changed.

Sampling: the assumption the document never states

A result describes what was tested. The leap from the tested portion to the whole lot is a sampling assumption, and it appears nowhere on most certificates.

For a homogeneous solution this is a small leap. For a dried solid it can be a large one. Materials can segregate during drying, during transfer, during transport: fines settle, particle sizes separate, moisture distributes unevenly across a tray. A sample from the top of a container can differ measurably from one taken from the bottom. Where a lot has been filled into many units, variation between units is a real quantity that a single composite result averages away.

A certificate can therefore be entirely accurate about the portion tested and materially unrepresentative of the container you hold. Nothing in the document will indicate this. The only mitigations are knowing how the sample was taken, which a thorough certificate states and most do not, and treating a single result from a large lot with proportionate caution.

Detection limits, and why "not detected" is not "absent"

Two thresholds govern what a method can honestly say, and conflating them produces a specific and confident kind of error.

The limit of detection is the smallest amount the method can reliably distinguish from noise. Below it, the method cannot tell presence from nothing. The limit of quantitation is higher: the smallest amount the method can measure with acceptable precision. Between the two lies a band where a substance can be seen to be present but cannot be assigned a trustworthy number, which is the small late peak marked in the figure above.

So not detected means one thing only: below this method's limit of detection. It does not mean absent. It does not mean zero. A more sensitive method may find the same substance comfortably, and both results are correct. This is why a result reported as not detected is uninterpretable unless the limit is stated alongside it. "Not detected" with no limit given is not a result; it is a reassurance.

The same logic applies to any specification written as less-than. The useful information is the number after the sign, and whether the method was capable of seeing below it.

Who issued it, and what accreditation actually covers

A certificate from a seller and a certificate from an independent laboratory are different kinds of evidence, and the difference is structural rather than a comment on anyone's honesty. A first-party document is an assertion by an interested party. A third-party document is an assertion by a party with something to lose by being wrong.

Laboratory accreditation, where it exists, is more specific than it is usually taken to be. The relevant standard for testing laboratories is ISO/IEC 17025, and accreditation under it is granted for a defined scope: particular methods, for particular measurements, on particular classes of material. A laboratory accredited for one family of tests is not thereby accredited for everything it is capable of running.

The practical question is not whether the laboratory is accredited but whether the test on your certificate falls inside the accredited scope. Scopes are published by the accrediting body and can be checked against the certificate number. A document that names an accreditation without naming the body or the certificate number has supplied a logo rather than a credential.

Two further things are worth looking for and are routinely absent. The first is whether the method was validated for this material rather than merely run: specificity, linearity, accuracy, precision, and the detection limits discussed above are properties of a method applied to a particular substance, not of the instrument. The second is whether a measurement uncertainty accompanies the headline figure. Every measurement has one. A result of 99.1% with an uncertainty of plus or minus 0.5% against a specification of 98.5% clears the bar; the same figure with an uncertainty of plus or minus 1.5% does not clearly clear anything, and the certificate that omits the uncertainty has concealed the ambiguity rather than resolved it.

Eight recurring failures

These are the patterns that recur, in rough order of how often they are encountered rather than how serious they are.

A ten-point check

A practical order of operations. Any failure early on makes the rest academic, which is why the cheap structural checks come before the interesting analytical ones.

1. Lot number. Does the string on the document match the string on the container? Not the product code. If it does not match, stop here.
2. Who issued it. A named laboratory with a contactable address, or the party selling the material? Both are legitimate documents. They are not equivalent evidence.
3. Accredited scope. If accreditation is claimed, is the body named and the certificate number given, and does the test performed fall inside the published scope?
4. Identity and quantity both present. At least one technique addressing what the substance is, and at least one addressing how much, each named individually.
5. Method conditions. Wavelength, column, gradient, mass range. Their absence is itself the finding.
6. Purity and content distinguished. Is there a figure for net content, or only chromatographic purity? If only the latter, the mass balance is open.
7. The supporting measurements. Water, counterion, residual solvent. These are what turn a purity figure into an account of the container.
8. Limits and uncertainty. Is every less-than accompanied by the limit, and does the headline figure carry an uncertainty that still clears the specification?
9. Dates with conditions. Test date, retest date, and the storage conditions the retest date assumes. Then ask whether the material has actually had that history.
10. The trace itself. Ask for the chromatogram. Look at the baseline, the symmetry of the main peak, and whether anything was still eluting when the run ended.

Ten checks, and most certificates fail at step two, step five or step six. They fail quietly, because a confident document with a high number on it invites you to stop reading at the number. The steps that catch the most are the structural ones at the top, which require no analytical knowledge whatsoever.

What a certificate cannot do at all

Being clear about the boundary is as useful as being clear about the contents. A certificate is silent on all of the following, and no amount of reading it more carefully will change that.

• Whether the document is genuine. A PDF is trivially editable, and a convincing one requires no skill. Confidence comes from the issuing laboratory being independent and contactable, and from the result being reproducible on request, not from the file looking official.
• Whether the sample represented the lot. Results describe what was tested. If the sampling was unrepresentative, the certificate is accurate and misleading simultaneously, and nothing in it will say so.
• What happened after testing. Everything between the laboratory bench and your hand is outside the document's scope: transport temperature, time on a dock, repackaging, the headspace of a container that has been opened. This is usually where material is actually lost.
• Whether the specification was appropriate. A result that clears its acceptance criteria tells you the material met the standard that was applied to it. Whether that standard was demanding enough is a separate judgement, and the certificate cannot make it.
• Anything about suitability for any purpose. A certificate is a record of measurement. It confers no fitness for any application whatsoever, and any document implying otherwise has exceeded what the measurements can support.

None of this is an argument for distrusting analytical documentation. It is an argument for reading it as what it is: a precise, narrow, dated, conditional record, which is far more useful than the pass-or-fail stamp it gets mistaken for. A certificate that states its methods, its conditions, its limits and its uncertainties is a stronger document for admitting what it does not cover, and the habit of asking what a number excludes is the whole of the skill.