Every letter we have printed

Every method in common use estimates composition from a physical property and a model, and the models were validated in populations that differ from the ones being scanned. The measurement is a calculation and it is reported as an observation.

Mass loss of any origin is associated with density change in the older literature, which makes the untreated comparator essential and frequently absent. Without it the question of attribution cannot be approached at all.

Applying trial findings to material of unknown provenance adds an uncertainty that no statistical caveat in the paper covers. Your publication is unusual in making that point where it belongs, which is beside every extrapolation rather than in a standing disclaimer.

Freight is the confounder nobody controls for. Two vials from the same lot, one shipped in January and one in July, are not the same sample by the time they reach an instrument, and a blind programme that ignores the season is measuring the courier as much as the manufacturer.

Two blind determinations from the same lot at two different laboratories is worth more than four determinations of four lots at one. Method variance is the largest term in most of these datasets and almost nobody designs to measure it.

A buyer sees a badge as a promise about the vial arriving next week. A service sees it as a receipt for work performed. Neither party is being unreasonable and the artefact sitting between them supports both readings.

Interval and half-life are related by arithmetic that is published for each compound in this class, and the discussion proceeds mostly without it. The accumulation ratio for a given interval is calculable and rarely calculated.

Extending an interval beyond the printed one is the departure nobody objects to and it deserves the same scrutiny as shortening it. A longer interval is safer with respect to tolerability and it also means a longer period at a dose that may not be doing much. There is a cost on both sides.

You describe the plateau as an energy-balance event and dismiss receptor desensitisation. Is there not a third possibility — that adherence quietly falls off at around a year and the plateau is partly a behavioural artefact of the trial rather than a physiological one?

The most useful thing this department could publish is a plain statement of what is and is not known about in-use stability for this class, with the sources. The absence of such a statement is why the folklore has filled the space.

Light exposure gets one clause in your article and deserves more. Several of these sequences carry residues that photodegrade, and a clear vial on a bench under fluorescent light for a fortnight is a real exposure. Amber glass and a drawer are free.

Oxidation of methionine is the other route worth naming and it is driven by headspace oxygen and by trace metals. A vial filled under nitrogen is materially better protected, and whether the fill was under nitrogen is not something any certificate I hold discloses.

The material this publication covers has no label at all, which makes the whole comparison a read-across from a different product. Saying so plainly, as you do, is more useful than the comparison itself.

Your comparison of the printed schedules across agents shows how much variation there already is between products, which undercuts the idea that any one schedule is uniquely correct. It would be worth stating the doses in the table with the intervals rather than beside them.

Where the material is of uncertain content, an apparent plateau may be a supply event rather than a physiological one. That possibility exists in this market and in no clinical trial, and it deserves to be named in any coverage aimed at these readers.

The word plateau implies a stable state and what the data usually shows is a decelerating one. Those are different shapes with different implications for what happens next, and the distinction is lost in every summary I read.

A plateau is defined differently in every paper I have read and not defined at all in most general coverage. Before arguing about whether one occurs, somebody needs to say over what interval and by what margin.

Supply interruption is the underdiscussed maintenance risk. A plan that depends on continuous availability of a specific product is a plan with an external dependency, and the last two years have shown how that dependency behaves under stress.

Nothing in the published evidence supports or refutes indefinite continuation, because nobody has run a study of that length. Where the honest answer is that the question is open, saying so is not a failure of reporting.

Blinding a withdrawal is harder than blinding an initiation, for reasons that are obvious once stated: the effects being withdrawn are perceptible. Any withdrawal trial reporting successful blinding should be asked how it was assessed.

The clinical trials that inform this discussion ran for a fixed period and stopped. Every statement about maintenance beyond that horizon is extrapolation, and the confidence with which extrapolations are stated in general discussion is not warranted by their source.

A note on vocabulary from a reader who works with long-term conditions. In every other context, maintenance describes a phase of ongoing management with regular review. Here it is used to mean the part where nothing needs attention, which is close to the opposite.

The ratio everybody quotes was measured in a population with a mean age and a mean starting composition, and it is applied without adjustment to people well outside both. Every term in that extrapolation is doing work that the original authors would not endorse.

Age and baseline muscularity change the answer more than most sources admit. A finding averaged across a trial population is not a prediction about a person at either end of that distribution, and popular summaries convert averages into individual expectations almost automatically.

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.

The population most at risk on this endpoint is the one least represented in the trials, which is a general feature of trial recruitment and a specific problem here. Extrapolating a bone result across age bands is not supportable.

Site matters enormously. Spine, hip and forearm measurements diverge in the same individual over the same period, and a single reported figure without the site is close to uninterpretable.

Specifications that are set after the result is known are not specifications, they are descriptions. The tell is a limit that sits a fraction below the reported value on every lot. Where a company publishes one limit and holds it across a catalogue, it is telling you the number was decided in advance.

I would like the trade to adopt a convention on significant figures. A specification of not less than 98 per cent and a result of 98.04 per cent are a pass by two hundredths of a point, which is well inside the repeatability of the method. The certificate should carry the method repeatability alongside the limit, or the pass means nothing.

You put every documentary finding to the company before publication, which is admirable and also means you are letting companies that do not reply escape coverage entirely. Silence should have a cost.

The lot number should appear in a form that can be matched to the vial without interpretation. Half the certificates I hold carry a batch code with a different number of digits from the one printed on the label, and reconciling them is guesswork the buyer should not be doing.

You describe the appearance line as decorative. It is not. Colour, cake structure and collapse are the only observations recorded before anything is dissolved, and a collapsed cake with an intact vacuum is a real finding about the lyophilisation cycle. It is the cheapest line on the page and one of the few that is not a number.

Half-maximal values from different assay formats can differ by an order of magnitude for the same ligand, which makes any table without the assay method close to decorative.

Constitutive activity of the receptor is measurable and rarely reported, and it changes how a partial effect should be interpreted. It is one of several parameters that would need to be held constant for any cross-paper comparison to mean anything.

Adding glucagon agonism to increase energy expenditure is elegant and it is also the limb that has to be balanced most carefully, because the same receptor raises hepatic glucose output. The ratio between the components is the whole design problem and it is quoted far too casually.

As a community pharmacist I would add one thing to your section on missed doses: the label window matters less than whether the patient then double-doses to "catch up". I have seen that twice this year and both times the patient believed they were following instructions.

Aggregates are invisible in the sense that matters: they are still the right molecule, so identity confirms, and they may be excluded from a reversed-phase separation entirely by never coming off the column. Whatever stays on the stationary phase is not in the denominator.

A modest addition to your list: anything present below the detection limit. It is obvious when stated and it is routinely forgotten, because a clean spectrum reads as a clean sample. The absence of a peak is a statement about sensitivity before it is a statement about composition.

Water is the invisible component nobody costs. A lyophilised cake that has taken up moisture in a warm warehouse weighs more and delivers less, and the purity figure is entirely unaffected because it is a ratio of peak areas. Two vials, same certificate, different dose.

Laboratories are accountable to whoever pays them, which is ordinary commercial life. What is unusual here is that the person relying on the report is almost never that person, and there is no mechanism by which they become one.