On this page
Check the input identity
Name the common horizon, unit, sign, exposure basis and estimation status. Ensure matched loss/probability records, a valid total and a supported joint model if combining components.
Check the threshold calculation
Order losses ascending and add probabilities for identical amounts. Apply the stated cumulative rule, keeping equality at a discrete boundary when the lower quantile is used. Retain the sample convention when comparing estimates.
Reconcile inclusion and average
| Layer | Question |
|---|---|
| Mass | Does the tail sum to 1−confidence? |
| Bounds | Is each inclusion within available probability? |
| Ordering | Were worse losses selected first? |
| Average | Was selected weighted loss divided by selected mass? |
| Events | Were strict and inclusive exceedance kept distinct? |
| Scope | Were forecasts, maximums and model summaries separated? |
Choose a correction tied to the error
A unit mismatch calls for a new input record; a boundary error calls for a cumulative/allocation reconstruction; a dependence gap calls for a joint model. Repeating the same calculation cannot restore missing evidence.