Validation Tools Taxonomy

What needs validating in a pharmacometrics activity, and the tools for each

Index of the ai-validation project: a taxonomy of the kinds of validation a pharmacometrics activity needs, the checklists and tools that exist for each, the gaps, and a validation plan that applies whether an agent or a person did the work.

A pharmacometrics analysis can go wrong at the question, its causal assumptions, the sources, the data, the code, the model, the predictions, the conclusions or the way it is presented. Checklists exist for several of these, among them the validation checklist written with Oliver Sander, the R coding checklist, and the uncertainty checklist. This project lists every kind of validation an activity needs, maps the existing tools onto them, and ranks the gaps. The same validation plan is meant to apply whether the work was done by a coding agent, an intern or a senior associate. Nothing is settled yet.

Documents

• Working specification. The taxonomy of thirteen kinds of validation, the tools and gaps for each, a validation plan template, the gaps ranked, and the open questions.
• Validation taxonomy. Thirteen kinds of validation, from study design and assays to the use of the answer, with the guidances, papers, checklists and tools for each, and the software and other fields PMX can borrow from.
• Validating code written by an AI agent. The project’s first specification, now one entry of the taxonomy: which parts of the existing validation checklist still apply when an agent writes all the code.