Parallel Run
Running the old and new systems side by side on the same inputs and comparing outputs, before the new one is trusted.
The most effective verification technique available for replacing a system whose behaviour is not fully documented — which is most of them.
The shape: both systems receive the same input, the old one's output remains authoritative, the new one's output is recorded and compared. Discrepancies are investigated, and each one is either a bug in the new system or an undocumented rule in the old one. Both discoveries are valuable, and the second is the one that makes this worth the cost.
Practical requirements: the new system must be side-effect free during the shadow period (no emails sent, no payments made), the comparison must tolerate legitimate differences such as timestamps and IDs, and there must be a defined exit criterion — a discrepancy rate below some threshold sustained over some period — or the parallel run continues indefinitely because nobody is willing to call it.