In the StrangeLoop FoundationDB talk, Will Wilson mentions the need for re-running simulations with the same deterministic seed and comparing the execution in order to detect sources of nondeterminism which have sneaked into the application.

It would be good to support this, if even at a basic level. One idea that has been floated is to use Tracing to collect a globally ordered set of execution events. This could serve as the basis for comparison to ensure that for multiple runs of the same seed, execution histories are identical.