Just IMO the 'big thing' is getting the alpha correct (as in: not overfit crap because you tested on same data multiple times, not ignoring physical constraints, nor returns boosted by some bug). In fact, having an idea of what your alpha/reason to get paid by market is going to be before you...