Achieving min cycles on a tournament puzzle is always challenging.  Despite containing no overtly nasty tasks, this one is no exception to that rule.  L=2 on both reagents, very little spare latency on the non-final inputs, and a fair amount of required duplication.  I had to tinker for an entire 6-hour flight before I saw the victory screen with 31 cycles.

It seems to me that the two general approaches would be to prebuild the central trapezoid, or buffer 9 singletons early in the process.  I'm betting that buffering will score better, since atoms leave the board faster and each task can be run on a 4-cycle loop instead of squeezing with a 2-loop.  

(But I actually just didn't want to try prebuilding, so wizards can feel free to prove me horribly wrong.)

No bloody clue how to build for X secondary, so I just tweaked bits around the edges and iterated on the lineages that got smaller numbers.  Really it's just the singleton buffering that changed--almost all else is identical to my first success. Didn't have the stomach to look for a redesign.