On this page
Compute the full-model mean
The original −10/0/10/50 losses with 30/40/25/5 percent probabilities have mean −3+0+2.5+2.5=2 loss units. This averages every model outcome and includes the gain outcome.
Compute the selected 90% tail average
The worst 10% mass has weighted loss 0.05×50+0.05×10=3. Dividing by 0.10 gives 30. Dividing by one instead would leave an unnormalised tail contribution, not the selected-tail average.
Use another original model to check the distinction
For losses 0/100/1,000 with probabilities 99/0.9/0.1 percent, mean loss is 1.9. At 95% confidence, exact-tail ES is 1.9/0.05=38 because the selected tail adds zero-loss mass to the positive-loss events.
Retain both if they serve the task
A tool may display mean and ES together, but each needs its own label. Changing one to match the other would discard the distributional question the reader is trying to answer.