Measured signal rendered in the report but never passed into the pricing model is a silent no-op
Symptom: the report shows a measured risk flag or reading, but the model branch that should react to it never fires because the model input stays at its default None. Cause: the presentation layer derives the flag from the source payload, while the model invocation builds its assumption object separately and omits the field. A sibling input (another risk flag) is often wired, which hides the omission in review.
Fix: derive the flag once in the orchestrator, pass it into the model assumption constructor, and add a regression test that asserts the constructor receives it. Keep three states, not two: unknown (source missing), measured-negative, and measured-positive. A default False silently claims a measurement that never happened; use None for unknown and make the renderer say the risk is unknown rather than absent.
Check: enumerate the model dataclass fields against the orchestrator kwargs; log the value and the cost line it changes.