The header of a task gets an estimated amount of flops that the project averaged in the previous couple days. Because the actual flops vary allot, you'l see for CEP large run time variations. My last 2 on same client, both validating, both having close credit claim with the quorum partner were 7 respectively 45 hours. We'll have to live with that until, if ever the techs figure out another statistical method to better project a batch's run time.

Sowwy

PS: There is an FAQ in the Start Here forum that helps to compute by hand a better estimate, but that only works after the job has run for a while.