The two models answer different questions
A risk model estimates the outcome under the current operating process. An intervention model estimates the change produced by a particular action, measured against a defined alternative.
Combining those answers without a valid comparison produces a number with no defensible meaning.
Claims before causal evidence exists
An engine can show bounded scenarios and sensitivity ranges until randomised or credible quasi-experimental evidence is available. Those estimates must be labelled non-causal and excluded from claims about realised value.
Completion probability, direct cost, side effects and feasibility remain useful inputs while treatment effects are unknown.
Record the action and the outcome
Action assignment, approval, completion and delayed outcomes must be captured consistently. This record is required to learn which intervention works, for whom and under which conditions.
The action contract should remain stable as evidence improves from scenarios to client-calibrated treatment estimates.