This reminds me very much of manufacturing quality testing, where you do initial DoE and validation builds to develop knowledge of the process. Eventually the knowledge of procc behavior at boundary conditions is known well enough that you can start to use alternative process monitoring approaches, like building a representative mockup of in-vitro conditions that represents the actual use conditions.

It doesnt have to be the same vector, just the same eigenvector.