How to Read a Peptide COA: Spotting Real vs. Fake Documentation

Cross-checks that need only the document and a calculator

Fabrication and recycling are failures of consistency rather than failures of completeness. Anyone assembling a convincing-looking document can copy a field list from a real one. What is much harder is making every derived quantity agree, because several of the numbers on a certificate are not independent of each other. The sequence determines the molecular formula. The formula determines the theoretical mass. The theoretical mass constrains what an instrument could plausibly have observed. Wherever those links break, the value was typed rather than measured, and that is true whether the cause is a deliberate edit or a tired analyst copying the wrong row.

The first link is sequence to formula. For a linear peptide the formula is the sum of the constituent amino acids minus one water molecule for every peptide bond formed, so a chain of n residues loses n minus 1 waters relative to the free amino acids. Common modifications shift it in known directions: a C-terminal amide replaces a hydroxyl with an amine group, an N-terminal acetyl group adds two carbons, two hydrogens and an oxygen, and forming a disulfide bridge removes two hydrogens. Nobody expects a receiving lab to do this by hand. Any peptide mass calculator will return a formula and both mass figures from the one-letter sequence in a few seconds. The point is not to reproduce the laboratory's work but to confirm that the formula printed on the page belongs to the sequence printed three lines above it. A formula lifted from a different compound is one of the most common artifacts of a reused template, and it survives because almost nobody checks.

The second link is formula to theoretical mass. Certificates often quote an average and a monoisotopic value, and the sibling walkthrough covers which is which and when each is appropriate. For consistency purposes the question is narrower: does the stated theoretical value follow from the stated formula, and if two values appear, do they differ in the expected direction and by an amount that scales with molecular size. A theoretical mass matching neither calculation is a transcription error at best.

The third link is theoretical to observed. An observed value should sit close to theoretical, and where it does not, the gap should be explainable as a recognized adduct, a water loss, an undeconvoluted charge state or a counter-ion being counted in. Two patterns deserve a question: an unexplained gap that matches no familiar shift, and an observed value reported as exactly equal to theoretical out to more decimal places than the instrument could resolve. Real measurements scatter. Perfect agreement is a sign of a calculator, not a detector.

The remaining checks need no chemistry at all. Dates should run in a possible order, with the fill or manufacture date first, the test date after it and the report date last. A test date preceding the fill date describes material that did not yet exist. And the lot string on the document has to match the lot string on the vial character for character, because that single field is what connects a page of numbers to a physical container.

Internal-consistency checks, in the order they cost least to run

Run them cheapest first. The lot comparison and the date order take under a minute and catch a surprising share of problems on their own. Only then is it worth opening a mass calculator, and only then is it worth zooming into the chromatogram.

A failed check is a question, not a verdict. Transcription errors are ordinary, they happen in well-run laboratories, and they are usually acknowledged and corrected without argument once pointed out. What separates a clerical slip from something worse is not the first failure but the response to it, and that is the subject of the last section here.

When the chromatogram and the summary table disagree

Most certificates carry two representations of the same analytical run. One is machine output: the trace itself, usually with an integration table beneath it listing each detected peak with its retention time, area and area percent. The other is a human-typed summary, the results block near the top that states a purity figure and often a retention time. The typed layer is where edits are easy. The trace is far harder to alter convincingly. Disagreement between the two is therefore informative out of all proportion to how small the discrepancy looks.

Retention time is the quickest of these to test. The summary quotes a time for the main peak; the chromatogram plots it. They should agree to the precision printed, because they describe the same peak in the same run. If the summary says one figure and the apex on the trace sits somewhere else, the summary was written against a different chromatogram. Note that agreement between two separate runs is a different question entirely, since retention drifts with column age, mobile-phase preparation and temperature; the check here is within a single document.

The integration table supports an even simpler test. Area percent is each peak's area divided by the total integrated area, so by construction the column sums to one hundred, allowing a unit or two in the final decimal for rounding. If the listed peaks add to noticeably less than one hundred and there is no line accounting for the remainder, peaks were excluded from the calculation. Exclusion can be entirely legitimate, most often for material eluting at the solvent front, but a complete report says so. If the headline purity in the summary block differs from the main peak's area percent in the table below it, the two numbers were produced independently, and only one of them came from the instrument.

The method description and the trace should also agree with each other. A time axis ending well before the gradient described in the method has finished means the image was cropped, which quietly removes any late-eluting peaks from view. A trace with no visible time axis cannot be checked against anything at all.

The strongest link between a chromatogram and your material is the sample identifier in the instrument report header, which normally carries the lot string or a work-order number the laboratory can tie back to it. When a chromatogram appears as an image cropped to just the plot area, that header is the thing that has been removed, and with it goes the only evidence that this particular trace belongs to this particular lot. The trace may be perfectly genuine and still describe someone else's material.

One last comparison costs nothing. Read what the summary says about peak shape, then look. A document describing a single symmetrical peak while the trace shows a clear shoulder or a second resolved peak has a summary that was not written from that trace.

Disagreements between the typed summary and the machine output

When any of these fail, the request that resolves it is specific: the native instrument report for that lot, exported as a full-page file rather than an image, with the sample header and both axes intact. That single file settles retention time, integration total, method agreement and lot linkage at once.

A supplier who can produce it usually does so quickly, because it is a routine export rather than a favor. A supplier who cannot has told you something useful about how far its documentation actually reaches back toward an instrument, without anyone having to make an accusation.

Fingerprints a reused template leaves across a document set

Building one convincing report and adapting it is far cheaper than fabricating each one from scratch, so recycling is the pattern most often encountered. The signals fall into two groups: things that ought to differ between compounds and do not, and things that ought to differ between runs and do not. Neither group is visible in a single document. They only appear when several documents from the same source are laid side by side, which is why the useful habit is to keep every certificate a supplier has ever issued to you in one folder.

Detector baseline noise is the most decisive of these, and the reason is statistical rather than forensic. The small wander of the signal between peaks is stochastic. Two genuine runs, even on the same instrument with the same sample minutes apart, never produce an identical noise pattern any more than two recordings of the same silence produce identical waveforms. If the wiggle between peaks is identical point for point across two reports, those are the same image with different text over it. Detecting this needs no software: open both files at the same zoom and flip between them, and an identical baseline snaps into place while a genuine pair visibly shimmers.

Retention time carries similar weight when it repeats across different compounds. Two unrelated peptides can coincidentally elute at similar times under a shared method, so a rough match means little. An identical value to two decimal places across several different compounds does not happen by chance at that precision.

A purity figure that is always the same round number is a softer signal with several innocent explanations. A supplier may be reporting against a threshold rather than a measurement, in which case the number is a specification and not a result. It may be rounding aggressively. Or the figure may simply not be generated per lot. None of those is fabrication, but all of them mean the number tells you less than it appears to, and the distinction is worth asking about directly.

Axis behavior is a quieter tell. Chromatography software normally autoscales the signal axis to the largest peak, so the axis maximum tends to differ from run to run. Identical axis extents across reports for different compounds suggest one underlying plot. The same reasoning applies to an analyst signature that appears as an identical image at an identical position, to report timestamps that are the same to the minute across lots produced weeks apart, and to a run of lot numbers incrementing by one with results that do not move at all.

Fairness matters here. A shared layout, a shared logo, a shared method for a family of related compounds and a shared certificate structure are all normal and desirable, and finding them proves nothing. It is identical measured output under different labels that has no innocent reading.

What to compare across two documents from the same source

The practical consequence is that document review gets better with volume. The first certificate from a new source can only be checked internally. By the third or fourth, cross-document comparison becomes available and it is considerably more powerful, because it tests the one thing a template cannot fake, which is that independent measurements produce independent output.

Keeping the original files rather than printouts matters for the same reason. A printed and rescanned page destroys exactly the fine baseline detail the comparison depends on.

File-level signals that cap what a document can support

Before any number is read, the form the document arrives in sets a ceiling on what it can be used to claim. A full instrument report is a recognizable artifact. It has a header block naming the sample, the method, the column, the instrument and the operator, with a timestamp. It has the trace with both axes labeled and scaled. It has the integration table. It usually has page numbering that shows how many pages the complete report ran to, which immediately tells you whether you are holding all of it.

A low-resolution image cropped to the plot area has none of that. It may well be a genuine picture of a genuine run, but it has been stripped of everything that would let a reader connect it to a lot, a method or a laboratory. Treating it as equivalent to a report is the substitution that most weakens a receiving check.

Missing axis information deserves its own mention because it is easy to miss. Without a labeled time axis there is no way to test the summary retention time. Without a labeled signal axis there is no way to judge whether small peaks were suppressed by scaling or genuinely absent. A plot with unlabeled axes is a picture, not data.

An absent method section has a similar effect. Column chemistry and dimensions, mobile phase, gradient, flow rate and detection wavelength are what make a purity figure reproducible and comparable with anyone else's figure for the same material. Without them the number cannot be checked by a second laboratory, and any disagreement between two labs becomes unresolvable. The same goes for an unidentified instrument and an unnamed analyst: those fields are what put a person and a machine behind the result.

A scan of a scan is worth noticing without being read as intent. Repeated generations of scanning introduce rotation, speckle and compression artifacts, each pass losing detail. It is often just an office with an old workflow. But it does conveniently destroy the fine baseline structure that cross-document comparison relies on, and it reduces a document to something whose provenance cannot be examined.

File metadata belongs in the same category of soft signals. Most PDF viewers show a document properties panel listing creation date, modification date and the application that produced the file, sometimes with an author name. A report generated by instrument or laboratory-information software looks different from one assembled in a word processor or an image editor. A creation date long before the claimed issue date is odd. A creation date after it is completely ordinary, since re-exporting, merging and redacting all reset it. Metadata is trivially editable and frequently wrong, so it is a prompt for a question and never evidence in itself.

Two more take seconds. Try to select the purity figure as text: genuine instrument exports usually have a text layer, while a page that is entirely a raster image has been printed and rescanned or built as a picture. And zoom in on individual numbers. A figure sitting on a slightly different baseline, or in a marginally different weight from its neighbors in the same table row, is a field that was overwritten.

How the delivery format limits the claims a document can carry

None of these on its own says a document is false. Small laboratories issue plain reports, and older scanning workflows are common in places that do careful work. The right conclusion is narrower and more useful: each missing element removes a specific claim the document could otherwise have carried, and what remains is what you are actually entitled to rely on.

Written that way the finding is easy to act on. Rather than deciding whether a document is trustworthy in the abstract, you end up with a short list of things it cannot establish, and each of those converts directly into a request.

In-house reports, third-party reports, and a lab merely named

Three quite different artifacts routinely travel under the same phrase, and the gap between them is larger than the wording suggests.

The first is a quality-control report issued by the manufacturer on its own letterhead. This is the ordinary case and most certificates in circulation are of this kind. It is self-reported, which is a real limitation, but it also has a genuine strength that gets overlooked: the manufacturer has the most direct knowledge of the lot, tested material it actually holds, and can tie the result to a fill record. A complete in-house report from a manufacturer that keeps records is worth considerably more than a thin document from an outside name.

The second is a contract-laboratory report supplied to you by the seller. This adds an outside measurement, which is worth having. What it does not add is independence of sampling. Standard contract-laboratory wording notes that results apply to the sample as received, which is honest and precise: the laboratory characterized whatever arrived in its sample vial. Who selected that sample, how it was drawn from the lot and whether it is representative are all outside the laboratory's knowledge and outside the scope of its report. The independence attaches to the measurement, not to the chain in front of it.

The third is a document that names a laboratory without evidence of engagement. A line reading tested by, or a laboratory logo placed on a manufacturer's own form, with no report number, no laboratory letterhead, no report of its own and no way to reach the named party. This establishes nothing whatever, and it is worth being blunt about that, because it visually resembles the second case closely enough to be mistaken for it.

Confirming that a laboratory actually issued a report is more approachable than it sounds. Genuine reports normally carry a report or work-order number and a named client, which exist precisely so the laboratory can retrieve the record. Ask the supplier for that number and for the laboratory's own contact details, then contact the laboratory quoting the report number and the lot, and ask one narrow question: did you issue this report, and is the copy attached complete. Practice varies on the answer. Some laboratories confirm authenticity readily, others decline to discuss a client's work at all, and a refusal to comment is not a signal about the supplier. What is a signal is there being no report number to quote in the first place. Where the laboratory holds accreditation, the accrediting body publishes its scope, and checking whether the test in question falls inside that scope is a public, free lookup.

A last point on limits. Independence never reaches further than the sample. An outside laboratory can report on what it received; it cannot certify a lot it never saw in bulk, and it cannot say that the vial in your freezer came from that lot. That link is carried by the lot marking and the custody record, not by the letterhead.

What each kind of document does and does not establish

Ranking these is not about preferring outsiders on principle. A detailed in-house report naming the instrument, the method and the analyst, with a full chromatogram and a lot that matches the vial, supports far more weight than a cropped image bearing an outside name.

The ordering that matters is by how much of the chain each document actually covers, and by how much of it can be checked by someone other than the party that benefits from it being true.

Escalating a failed check without guessing at motive

The first move after a failed check is procedural rather than investigative. Mark the material as unresolved and stop drawing from it. A discrepancy found before the material enters an experiment costs a few emails; the same discrepancy found afterward costs the experiment, and every result that depended on it.

Then record what you saw, verbatim and immediately. File name, page, field name, both conflicting values as printed, and the date you looked. Save an untouched copy of the original file somewhere it will not be overwritten. If a corrected document later replaces the original, your record is the only remaining evidence that the discrepancy ever existed, and without it the conversation becomes a matter of recollection.

Ask narrowly and factually. A message reading that the certificate for a given lot lists one formula and a sequence that does not correspond to it, and asking which is correct for that lot, gets an answer. A message implying the document is fake gets a defensive non-answer, and since the most common cause by a wide margin is a transcription error, an accusatory opening converts a routine correction into a dispute for no gain. The tone that works treats the supplier as the party best placed to explain, because usually it is.

A good answer has a recognizable shape. It identifies which specific figure was wrong and how it got there. It arrives with a corrected document that carries a revision number, a revision date and a short statement of what changed, leaving every unaffected field untouched. Where relevant it comes with the underlying instrument output rather than another summary. A weak answer also has a shape: a fresh file with the disputed number silently altered and the original date retained, a verbal assurance with no document, a document for a neighboring lot, or a redirect to a general quality page that discusses nothing specific.

If the first exchange does not settle it, the requests that follow are concrete and ordinary: the native instrument report for that lot with the sample header and both axes intact, the laboratory's report or work-order number, and the fill date. Ask also whether a retain sample of the lot exists and who holds it, since a retain is what makes any later resolution possible at all.

When those are exhausted, independent re-testing is the only remaining resolution, and it helps to be clear about its reach before spending on it. It answers a question about the specific vials you send and nothing else. It does not validate or invalidate the supplier's other lots, it does not retroactively authenticate the disputed document, and it introduces a sampling question of its own, since a laboratory can only report on what it receives from you. Those are limits worth knowing in advance rather than discovering in the report.

Escalation sequence for a document that fails an internal check

The uncomfortable conclusion of all of this is that document review has a hard ceiling. Every check here tests whether a document is self-consistent and plausibly authentic. None of them tests the material. A certificate that passes every cross-check still says nothing about the vial in your hand unless the lot string on that vial matches the document and the custody between them is unbroken.

That is why the lot marking, the receiving record and the storage log carry more weight than any single field on a certificate. A flawless document attached to an unlabeled vial, or to one whose lot was never recorded on arrival, is a well-written statement about material you can no longer identify.

Questions this guide gets asked

Where to read next

• How to review a peptide certificate of analysis the field-by-field anatomy this guide assumes you have read
• Reading an HPLC chromatogram: purity by area how integration produces the number the summary block quotes
• Mass spectrometry for peptide identity confirmation which theoretical mass applies and what tolerance is reasonable
• Lot traceability: matching label, COA and records the custody side, where document review runs out
• Third-party lab testing and certificates