What One Certificate Cannot Tell You: Lot-to-Lot Variation and Trending
Put four purity figures in production order: 99.4, then 99.1, then 98.8, then 98.6. Every one of them passes. Every certificate that carried them was unremarkable, and anybody reading those documents one at a time would have found nothing to comment on. Laid side by side in sequence, they say something none of them says alone — a process is moving. That gap between what one report can show and what a series shows is the subject here, and it is also the reason the most useful question anybody can ask about a supplier is the one a certificate is built not to answer.
Three shapes worth recognizing
| Shape | What the numbers do | Usual origin |
| Drift | Successive lots move steadily in one direction | Gradual change somewhere in the process |
| Step | Level shifts abruptly and then holds | New raw material, resin, column or operator |
| Widening | Average stays put while scatter grows | Loss of control; reaches the limit before the mean does |
Each of the three is invisible from inside a single report, since each lot on its own is compliant and ordinary. And purity is rarely where the signal appears first. Impurity entries trend just as informatively and usually sooner — a deletion sequence walking from 0.2% up to 0.6% over five lots describes a synthesis step whose efficiency is slipping, of the kind covered in deletion, truncation and oxidation impurities.
Normal and acceptable are different questions
Plotting one attribute across consecutive lots, in the order they were made, is all that trending is. Do it consistently and a run of forgettable individual numbers resolves into a shape: holding, drifting, or spreading out.
Add control limits and it becomes formal. Those limits come from the process’s own history, and emphatically not from the specification — which is the entire point of the exercise. Acceptability is what a specification defines. Normality is what control limits define. A result that clears the specification while falling outside the range the process usually occupies is a signal of the sort treated in out-of-specification results and retesting, and nobody holding that lot’s certificate will ever see it.
One number, two sources of scatter
Behind every reported value lies a distribution, and it has two contributors. The process varies from lot to lot. The method varies from determination to determination. What gets printed is a single draw from the two combined, and the printed figure is silent about which contributed how much.
So 99.1% on one lot and 98.3% on the next could be a process swinging by the better part of a percent. Equally it could be a thoroughly stable process observed through a method whose uncertainty spans that distance, in the manner set out in measurement uncertainty. Disentangling them calls for the same material measured repeatedly, and that exercise never reaches a buyer.
Why the history cannot live on a certificate
Trending needs the record, and the record is the manufacturer’s property. It is not a certificate field, it is not a reasonable thing to request in that form, and no individual report is deficient for omitting it.
Worth saying outright, because the opposing idea — that a sufficiently good certificate would demonstrate consistency — mistakes what the document is for. A certificate that displayed a trend would be reporting on lots the purchaser never bought.
The partial series a repeat buyer can build
Buy the same item more than once and something resembling a trend becomes available, with genuine caveats attached. Keep the certificates: three or four for one item, filed by date, will show whether the reported figures are moving, and the filing involved is the ordinary discipline described in inventory labels and records. Compare only comparable things, because a value produced by a different laboratory or a different method is not another point on the same series — that being precisely the difficulty method transfer exists to handle. Read past the headline figure to the impurity rows, which move earlier. And keep expectations proportionate: a difference living in a digit the method cannot resolve reflects a rounding convention rather than a change in the material, as significant figures and rounding explains.
Two things this cannot do. It cannot tell process variation from method variation. And it never includes lots that were not sold, which makes a buyer’s series a filtered version of the manufacturer’s — filtered, moreover, by something that is not chance.
Proxies for consistency, ranked honestly
- Lot identifiers that increment coherently, in the sense described in batch and lot numbering, indicate that lots are genuinely being tracked.
- A certificate format that stays stable, carrying the same method parameters from one lot to the next, indicates the measurement is not being reinvented each time.
- Measured values in place of verdicts indicate there is anything to compare at all.
These are weak signals taken individually, and they are also what is on offer. Meanwhile the certificate itself remains exact about a narrow thing: what one method found, in one sample, from one lot, on one day. It does not promise that the following lot will match, nor that the remainder of this lot matches the portion sampled — the question posed in sampling plans and batch representation — nor that nothing has shifted since, which falls to the stability protocol. That specificity is a strength rather than a shortfall. The strong signal, the trend, stays with whoever made the material, and the sensible stance for a buyer is to know it, keep a series of their own, and read each certificate for the narrow thing it is.
