We explain the concepts and application of conditional power, conditional error and predictive power.
New here is the application to non-proportional hazards problems, which creates challenges beyond these calculations when the underlying treatment effect is assumed to be constant over time.

Problems we wish to solve under general assumptions including non-proportional hazards include:

1. Conditioning on a single interim test (Z-value or observed treatment effect), what is the probability of:
   a) A single future test crossing some future bound tested by including the original data and incremental data
   b) Any of multiple future tests crossing a group sequential bound.
2. Assumptions about future treatment effects are key.
3. Averaging conditional power over a posterior distribution would be helpful