Hello seah0210,
Unfortunately proteins don’t care about our target resolutions… many aspects of your data (from the protein intrinsic nature to the data processing, passing by quality of freezing, data collection…) will impose that limit.
On box size, I’ve seen both situations, a few where a tight box improved resolution, and also the opposite. That is something to be optimized according to the situation - doable with a small subset of particles.
About particle selection, depending on the final numbers yes, you can select only the best-looking classes, just keep in mind that this usually means you are strongly biasing the final particle set - always check the orientations plots after refinement.
If you haven’t already, take a look at this trend: https://discuss.cryosparc.com/t/cryosparc-workflow-discussion-comparing-different-processing-orders/16320/3
Luck