Which single measurement would I keep if I could only have one?

I am comparing a supplier certificate against an independent result on the same lot.

The failure mode I am trying to avoid is making this decision emotionally.

I have twelve months in view and I would like the plan to survive that long.

How would you structure this, and what thresholds would you set in advance?

The single most misleading statement on a research-grade certificate is a lot number with no statement of how many vials from that lot were tested.

Under AQL sampling plans, testing two vials from a fifty-vial lot gives you an operating characteristic curve that tells you what risks you are accepting.

Stratified sampling — testing one vial from the top, one from the middle, and one from the bottom of a shipment — is cheap insurance against segregation.

Published data on lot homogeneity from manufacturers who sample multiple vials consistently find variation below the published specifications, suggesting the sampling plans work.

One qualification: testing more vials gives better confidence, but at some point the cost outweighs the benefit.

Assume segregation is possible, and design your sampling to catch it if it exists.

Sampling plans exist precisely because testing everything is expensive, and they define the statistical relationship between sample size and lot-wide inference.

Published segregation failures show that even modern automated processes sometimes produce lots with measurable vial-to-vial variation.

For a quantitative result like content, the acceptable range determines how many vials you need to test to establish the lot complies.

Sampling plans for pharmaceutical manufacturing are defined in ISO 2859 and ANSI Z1.4, and they are based on statistical sampling theory.

The practical summary: a lot number without a sampling statement is a lot number without meaning.

edited 23 Apr 2026 by kwn_analytical — added the placebo-arm figures

A certificate that reports one test result on one vial extrapolates to claim that all two hundred vials in the lot are identical, which is an assumption worth questioning.

If the lot was manufactured in multiple batches, testing vials from each batch separately establishes whether batch-to-batch variation is acceptable.

The underlying point is that testing a vial that has been open in the lab for three months is testing aged material, not the fresh lot, and the result should be explicitly noted as a retest.

Lyophilised peptide homogeneity studies show that vial-to-vial variation is usually small but occasionally large, depending on the distribution in the freeze-dryer.

If testing multiple vials, state how many you tested and why you chose those vials.

Start from the question: how many vials from this lot do I need to test to claim that the lot meets specification, and the answer depends on both the lot size and the acceptable risk.

The sample size determination requires choosing a confidence level and an acceptable error rate, and the smaller the error rate you want, the larger your sample must be.

I would treat a "complies with" statement without sampling details as a claim rather than as evidence.

If you only pay for one test, pay for quantified content. Purity is the number everyone quotes and content is the number that changes what you do.

The practical consequence is that spot-testing one vial from a new supplier is better than assuming they are all the same.

If you have reason to suspect inhomogeneity — different appearance in different vials, or a long or warm shipment — testing more vials is the diagnostic move.

The ICH Q3A and Q3B thresholds for reporting, identification and qualification of impurities are the framework the pharmaceutical industry works to, and they are worth reading even though nothing in the research-grade supply chain is obliged to meet them, because they tell you which numbers a competent analyst would consider worth reporting at all.

In practice: ask for the chromatogram, check the method section, check the lot number against the vial, and set your accept threshold before you see the result rather than after.

edited 22 May 2026 by mz_4113 — added the method parameters