What One Certificate Cannot Tell You: Lot-to-Lot Variation and Trending

A certificate describes one lot. That is its scope and its honest limit, and it means the most useful question about a supplier — do their lots resemble each other — is the one question a single certificate is structurally incapable of answering.

A result is a point, not a distribution

Every measurement is drawn from a distribution with two sources of spread: the process, which varies from lot to lot, and the method, which varies from determination to determination. A single number is one draw from the combination of the two, and nothing in it distinguishes which contributed what.

Two lots reported at 99.1% and 98.3% might be a process that varies by nearly a percent, or a stable process measured by a method whose uncertainty is that wide, as described in measurement uncertainty. Separating the two requires repeated measurement of the same material, which a buyer never sees.

What trending is, and where it lives

Trending is plotting the same attribute across successive lots in production order. Done consistently it turns a series of individually unremarkable results into a picture: a process holding steady, a process drifting, or a process whose spread is widening.

The formal version adds control limits derived from the process history — not from the specification. This distinction is the whole point of the technique. Specification limits say what is acceptable; control limits say what is normal. A result inside the specification but outside the process’s own normal range is a signal of the kind discussed in out-of-specification results and retesting, and it is invisible to anyone reading that lot’s certificate alone.

The three signals a trend shows and a certificate cannot

• Drift. Successive lots moving steadily in one direction — 99.4, 99.1, 98.8, 98.6 — all compliant, all unremarkable individually, and together a process changing.
• A step. A sudden shift in level, usually traceable to a change in raw material, resin, column or operator. A single certificate on either side of the step looks normal.
• Widening spread. The mean holds but the scatter grows, which is a loss of control that reaches the specification limit before the average does.

An impurity profile trends as informatively as purity does, and often earlier: a deletion sequence creeping from 0.2% to 0.6% across five lots is a synthesis step losing efficiency, in the way described in deletion, truncation and oxidation impurities.

Why this is a manufacturer’s tool, not a certificate field

Trending requires the history, and the history belongs to whoever made the material. It is not something that appears on a certificate, is not something a buyer can reasonably ask to be printed there, and is not a gap in any individual report.

It is worth stating plainly because the alternative framing — that a good certificate should somehow show consistency — misdescribes what the document is. A certificate that reported a trend would be reporting lots the buyer did not purchase.

What a buyer can actually assemble

Someone buying repeatedly can build a partial version of the same picture, with real limitations:

• Keep the certificates. Three or four for the same item, in date order, show whether the reported figures move. The record-keeping this requires is the ordinary kind described in inventory labels and records.
• Compare like with like. A figure produced by a different method or a different laboratory is not a point on the same series — the same problem that method transfer exists to manage.
• Watch the impurity entries, not only the headline. They move first.
• Expect small movements. Differences confined to a digit the method cannot resolve are display choices, not findings, as set out in significant figures and rounding.

What this cannot do is separate process variation from method variation, and it cannot see the lots that were never sold. A buyer’s series is a filtered sample of the manufacturer’s series, and the filter is not random.

The limits of a single certificate, stated plainly

One certificate establishes what a method found in one sample from one lot at one time. It does not establish that the next lot will resemble it, that the rest of that lot resembles the sample — the question in sampling plans and batch representation — or that the material has not changed since, which is the province of the stability protocol.

None of that makes the certificate weak. It makes it specific, and a document that is specific about what it covers is more useful than one that implies more than it measured.

What consistency looks like from outside

Absent the history, the observable proxies are indirect and worth ranking honestly. Lot numbering that increments coherently, in the sense of batch and lot numbering, tells you lots are being tracked. Certificates in a stable format with the same method parameters from lot to lot tell you the measurement is not being re-invented each time. Reported values rather than verdicts tell you there is something to compare at all.

These are weak signals individually and they are what is available. The strong signal — the trend — sits with the manufacturer, and the realistic position for a buyer is to know that, keep their own series, and read a single certificate for exactly what it is.