Open this dialog from the Setup criteria button on the Sensitivity Specs Panel. It is used to specify when a sensitivity run will be terminated and discarded based on the value of the payoff exceeding the specified bounds. This is normally use to explore ranges of outcomes that might be realized given uncertainties around model parameters. To do this the payoffs chosen will be calibration payoffs and the value chose would be an increase from the best calibration payoff. This provides similar information to the confidence bounds option in the Optimization Specs. As a rule of thumb a value that is 2 bigger than the calibration payoff would give 95% confidence bounds for weighted squared errors (where 2*payoff is chi-squared with 1 degree of freedom). In addition to being multidimensional, this approach allows you to include both the calibration period and the future which show how much the calibration constrains future outcomes.
For each payoff you have defined (if you have only defined 1 payoff there will only be one choice), you can specify a value that is around 2 bigger than the best payoff achieved to get the 95% confidence bounds extrapolated into the future (assuming sum of squared errors).
When a run exceeds the specified threshold it will be discarded. Runs may finish early because of this.