Recomputing an outcome while one input at a time moves across its plausible range, with every other input held at its base value, to rank what drives the result.
Sensitivity analysis answers which inputs a result actually depends on. Starting from a base case, each variable in turn is moved to its low value and its high value while the rest are held still, and the outcome is recomputed each time, so the spread it produces is a clean measurement of that one variable’s influence. Holding the others still is what makes the results comparable, and it is also the technique’s limitation, because it can say nothing about variables that move together. The output is a ranking, usually drawn as a tornado diagram, and it separates the few variables worth managing from the many that barely register.
Where this comes up
One bar per variable, the widest at the top, all of them measured against a baseline down the middle. A tornado diagram ranks what could move an outcome, and by how much.
A risk might happen. An issue already has. That single difference decides which artefact a line belongs in, what you record about it, and who you have to tell.
Contingency reserve sits inside the cost baseline, for risks already named and priced. Management reserve sits above it, for the ones nobody named. Level and authority, not size.
Most risk registers record one number per row. The number that actually justifies spending money is the distance between two of them.