Human-relevant is the beginning of the question.

A model can contain human cells and still miss the biological interaction that determines a clinical outcome. Relevance must be demonstrated for a purpose.

Start at the decision

Are you evaluating irritation, liver injury, target engagement or a long-term treatment response? Different questions require different experiments, comparators and standards of evidence.

Do not confuse resemblance with prediction

More complexity is not automatically more useful. A simpler assay with well-characterized performance may support one decision better than a sophisticated model without appropriate validation.

Close the learning loop

Specify the expected signal, test the model, compare with independent evidence and revise the decision. That loop, rather than any particular device, is the core of the Velocity framework.