Know whether the method is precise enough for the limit it is judging
An agent that reads the qualification package, quantifies how far the development bench method sits from the validated GMP method, and finds every release result close enough to the specification that the method cannot cleanly call it.
The method qualified. Nobody asked what its precision does to the specification.
A method is qualified and put to work. Most parameters pass. Intermediate precision comes in above its criterion, an investigation opens, and the item sits on a list while the method is used anyway because there is nothing else to use. Separately, the development bench version of the method reads a little differently from the validated GMP version, which everyone knows and nobody has quantified.
Then a result comes in at 1.9 against a limit of 2.0 and passes. Whether that pass is real depends entirely on how much the method moves, and the number that would answer it is sitting in a qualification spreadsheet that nobody has connected to the release data.
Making that connection by hand means reconciling a working Excel file against LIMS, applying the precision figure across the whole release population, and doing it carefully enough that you would defend it. It is a two or three day job, so it happens when an inspector asks, not before.
The qualification package, applied to every result the method has produced.
The agent is given a method and its qualification workbook. It extracts every qualification parameter against its acceptance criterion, quantifies the bias between the development bench method and the validated GMP method on shared material, derives an uncertainty band from precision and bias together, then queries every release result the method has produced and returns the ones sitting inside that band of the specification limit.
Precision is not a number in a report. It is a width around every result you have ever released against.
Qualification is usually treated as a gate that a method passes once. The more useful reading is that qualification tells you how much the method moves, and that number has consequences for every result it produced afterwards. Connecting the two is arithmetic. It is only ever skipped because the two sets of numbers live in different places.
- The qualification workbook as it actually exists, which is usually an uncontrolled spreadsheet: accuracy and spike recovery, repeatability, intermediate precision, linearity, range, limit of quantitation and solution stability, each against its acceptance criterion.
- Paired-material comparison between the development bench method and the validated GMP method, on the reference standard, engineering material and retains that were run on both. Materials run on only one method are reported as unpaired rather than quietly dropped.
- An uncertainty band built from intermediate precision and observed bias as two separate components, with the arithmetic shown rather than collapsed into a single figure.
- LIMS for every release result the method has produced, joined to the specification, stratified drug substance from drug product, with results tied to a closed lab investigation excluded and reported.
- The results the method cannot cleanly resolve: every value sitting within the uncertainty band of the limit, listed by batch with its value and specification, which is the output that changes what you do next.
The agent quantifies what the measurement system can and cannot resolve. It does not say a result was wrong, and it does not decide whether the method is fit for purpose, whether the specification needs revisiting, or whether qualification can be signed. Those stay with an AD scientist and the quality unit. Where intermediate precision has itself failed, the agent says plainly that the uncertainty estimate is built on a failing parameter.
From a question you answer under audit to one you answer before filing.
- Qualification data in a working spreadsheet, release data in LIMS, joined by hand
- Bench-to-GMP bias known anecdotally and rarely quantified
- Precision applied to the population almost never, because it is tedious
- The question gets asked by an inspector rather than before the filing
- Qualification, bias and the full release population read in one pass
- The uncertainty band shown with its components, not asserted
- Cheap enough to re-run whenever the method or the specification moves
- An AD scientist decides what it means for fitness and for the spec
These figures come from my demo environment running on synthetic CMC data. They are not client results. The "today" column is what I watched teams actually do across my career; the "with an agent" column is measured on the demo, against a data model built to be representative rather than against your systems.
The qualification data has to be reachable, and it usually lives outside your validated systems.
Qualification and method development data sits in spreadsheets more often than in a system, which is not a problem in itself. The agent reads spreadsheets. It becomes a problem when the file lives on a personal drive with three versions, when the paired bench-and-GMP comparison was never run on shared material, or when the release results cannot be joined back to the method revision that produced them.
The Data Diagnostic tells you which situation you are in for the methods that matter, and what it takes to close the gap, before you commit to a build.
The Data Diagnostic, including the price
Your IT and quality systems group will want the architecture, the read and write paths, the Part 11 position and the validation approach. That is all written down on the governance and validation page, in a form you can forward or print.
Send me a workflow.
Tell me the process that eats your team's week. I'll record an agent running it on your data model and send it back. No call required.
kyle@kylelangham.comNo form, no gate, no email capture.
Would rather talk it through? 30 minutes on your workflow, no pitch.