Why sensitivity analysis matters
Sensitivity answers the planning question: which inputs matter most for this outcome, and by how much? Each score is the interventional effect of a one-standard-deviation shift in the source node, propagated through all downstream paths to the target. Unlike feature importance in black-box models, these scores carry a causal interpretation — they measure the effect of a do(), not a correlation.
Sensitivity analysis guides resource allocation. A high-sensitivity node that is tractable to change is a natural intervention target. A high-sensitivity node that is difficult to change tells you the system is largely outside your influence from the levers currently in the model. The comparison between nodes is only meaningful when inputs are standardized — which is why standard-deviation scaling is essential here.
Top weights ranked by sensitivity swing. Bars anchored at zero — positive weights: blue right / red left; negative weights: pattern inverts.