By the end of this phase you should be able to show the following.
| Area | What you have done |
|---|---|
| Technical performance | Measured your primary metrics on a held-out test set, compared them with current practice and checked calibration |
| External validation | Repeated the measurements on data from at least one other site, which shows whether performance carries over to new settings |
| Fairness | Reported performance for each relevant patient group and judged whether the differences matter clinically |
| Usability | Tested the tool with representative users and measured acceptance, task time and errors |
| Clinical utility and safety | Run the tool prospectively in practice (first in silent mode, then in an impact study, with a randomised trial where feasible) and recorded adverse events and near misses |
| Robustness | Tested performance under realistic variation and, where relevant, deliberate manipulation of inputs |
| Explainability | Checked that explanations match what the model does and that clinicians understand them |
| Reporting | Written up the results following the relevant international reporting guideline, including failures and negative findings |
External validation shows that performance transfers to other data. It does not show that patients benefit; only a prospective evaluation in clinical practice can show that.