Every letter we have printed

Substudy participants are recruited at a subset of sites, and site is a proxy for a great deal — population, practice, adherence support. Treating a substudy as a random sample of the parent trial is an assumption nobody tests.

Prespecification matters here more than usual. A composition endpoint added after the parent results were known is a different kind of evidence from one written into the protocol, and the papers do not always make clear which they are reporting.

What happens when a certificate turns out to be wrong is the question your article stops short of. In a regulated setting a corrected document is reissued to everyone who received the original. Elsewhere it is corrected for whoever complained, and the original stays in circulation. That asymmetry is worth a piece of its own.

The accountability gap is contractual rather than analytical. Nobody in the chain has warranted anything to the buyer, so a wrong figure has no consequence, and a system with no consequence for error will produce errors at whatever rate is convenient.

You describe reused chromatogram images as a tell. I work in a contract laboratory and I would note that some data systems export a representative overlay rather than the individual injection, so identical-looking traces can occur legitimately across a batch series. It is still worth asking about; it is not the smoking gun your article implies.

We started photographing the vial, the label and the seal before sending anything for testing, and it has settled two disputes since. It costs a minute and it converts a claim about provenance into a record of it.

Storage between arrival and submission is the step most blind programmes describe least. A vial that sat on a desk for a fortnight before going to the laboratory has had an experiment performed on it that nobody designed.

Community-funded testing has solved the money problem in a handful of cases, and it introduces a new one: the group choosing the target has its own preferences about the answer. It is a better arrangement than seller-funded testing and it is not neutral.

What worries me more than shading is selective publication. A seller who commissions four determinations and publishes the best one has not corrupted any laboratory. Every document is genuine. The set is the misrepresentation, and no laboratory is in a position to detect it.

An image is a poor container for a conditional statement. Everything interesting about a determination is conditional — on the sample, the method, the date — and none of it survives compression into a graphic. The format has selected the message.

Thank you for declining to tell readers what to do. The restraint is unusual and it is the reason I trust what you do print.

Residual solvent and counter-ion content are the other two systematic absences, and they interact with the first. A trifluoroacetate salt can carry a substantial mass fraction that is neither peptide nor water, and a buyer weighing the vial contents is weighing all three.

Your list of absences should include the negative result. Certificates report what was found and not what was looked for and not found. A line stating that no impurity above the reporting threshold was detected other than those listed is a much stronger statement than a table that simply stops.

Units are quietly inconsistent across this trade. Water content by weight and water content by the Karl Fischer determination are usually the same thing and occasionally are not, and the abbreviation on the page does not distinguish them.

A technical query on your identity row: you specify 4111.1 Da monoisotopic with a tolerance of ±10 ppm, which is 0.04 daltons. Is that not tighter than most contract laboratories will commit to on a peptide of that size?

An accreditation body will investigate a complaint about a laboratory within its scope. Almost nobody in this market knows that, and I would put the route in a sidebar: it is free, it is real, and it is used vanishingly rarely.

Reporting on management would be improved by a single distinction: approaches with randomised evidence, approaches with observational evidence, and approaches with none. Most articles mix all three in one list.

Management strategies circulate in this area with very little trial support, and the honest way to present them is as reported practice rather than as evidence. Your piece does that; a great deal of adjacent coverage does not.

The reporting asymmetry is worth naming. Severe events are newsworthy and mild ones are not, so the visible discussion is drawn from the tail of a distribution and read as though it described the middle.

Intake falls when total intake falls, which is arithmetic rather than a finding, and it is presented as a discovery in a good deal of coverage. The interesting question is whether it falls disproportionately.

Supplementation before a test produces a value that reflects the supplement rather than the status, sometimes for days. Any series that includes both supplemented and unsupplemented draws is not a series.

I stopped treatment fifteen weeks ago and my panel is worse than I expected. I had a panel at five weeks that looked fine and I had assumed I had escaped. Your point about the twelve-week timing was the explanation nobody offered me.

A blind purchase can still be detected. Unusual delivery addresses, first-time customers ordering a single unit of an unusual line, payment methods that stand out — a supplier paying attention has several signals available, and none of them require anybody to behave badly.

There is a survivorship effect in the other direction too. Suppliers who fail a blind determination and quietly disappear from the market never appear in anybody’s dataset, and the surviving distribution looks better than the underlying one ever was.

Time of day matters for several of the analytes discussed in this context, and a series collected at different hours is a series with a systematic wobble in it. Same laboratory, same time, same conditions is worth more than any additional test.

An optimal range is a different concept from a reference interval and the two are used interchangeably in a good deal of what circulates. Optimal ranges usually come from a source with a product attached.

Your piece assumes readers have a clinician ordering panels for them. A great many of us are buying our own, from services that supply an interval, a flag and nothing else. The apparatus you describe is not available to us at all.

Assay methods for several of these differ between laboratories more than for common chemistry, and the between-method variation can exceed the width of the interval. Comparing across providers is particularly unreliable here.

The archive is the asset. Being able to follow one question across three years of issues is a different experience from reading any single piece, and very few publications make it possible.

Sample sizes in the withdrawal arms are typically a fraction of the initiation cohorts, which widens every interval in the interesting part of the study. Coverage tends to quote the point estimate from the smallest arm in the paper.

The ethics of a withdrawal design deserve a sentence somewhere. Randomising somebody to lose an effect they have obtained is defensible and it is not trivial, and the consent process for these studies is more interesting than the results.

The published trials that continued longest are the ones least discussed, presumably because their results are the least dramatic. A quiet extension study is more informative about a long commitment than any headline finding from a twelve-month arm.

pH is the strongest lever on most of these pathways and it is set by a buffer the buyer did not choose and cannot see. The most important stability variable in the vial is undisclosed.

Hydrolysis at the peptide bond is slow at refrigerated temperatures and not at room temperature, and the difference across that range is much larger than people assume from the small numerical gap.

Mass spectrometric detection on the same chromatographic run gives you an orthogonality of a sort for nothing, because a co-eluting species of different mass will show up in the extracted trace even when the ultraviolet peak looks clean. The separation is not orthogonal but the detection is, and for the co-elution question that is what matters.

Two-dimensional liquid chromatography now exists in a form a service laboratory can afford, and it collapses the whole argument: the second separation happens on the fraction rather than on the sample. It is still a specialist purchase, but the price is falling and the case for it in this market is stronger than in most.

You write that only one laboratory attached its chromatogram to the private buyer report. That was probably us. We started doing it five years ago because the PDF seemed incomplete without it. It costs us nothing to add — the instrument generates it automatically — and it solves exactly the dispute-resolution problem you describe. More laboratories should do it, and the reason they do not is not technical.

On the section about diode-array detection and peak purity, I would add that true peak purity assessment requires library matching or at least spectral comparison across the peak width. A homogeneous spectrum tells you the peak is probably pure. A spectrum that shifts across the peak tells you it is not, and that information closes a gap the article identifies correctly.

The pieces I have found most useful are the ones that concluded nothing very much. Reporting the absence of a finding is a service and it does not read as one at first.

The most valuable custody artefact I have is a courier record showing a parcel held for six days at a hub in July. It explains a determination that would otherwise have looked like a manufacturing failure, and it cost nothing to obtain.