Thanks to Pibero for putting in a modification to the trigger simulator that made it really easy to get passes_L0() to work right on both real and simulated events.  The following shows the number of events at each stage for a 500 event sample of W's.

Without the L0 check:

With the L0 check (note that no W the number of events that passed the L2 check remains essentially the same):

This doesn't guarantee that the mapping of the trigger patches back into their mudst container is correct, but it seems unlikely that that's gotten mangled.  

So it looks like L0 can be implemented uniformly now.