Question about Super Res Gain reference handling

[R Nwo]

Hi, 

We have a data set that we are using for testing, collected on a Falcon 4i at pixel size 1.19 (4K resolution), or 0.595 (8K super resolution). 

In previous instances of WarpTools (dev33 or below), we were able to do motion correction at 8K, and Fourier crop to 4K using these parameters for preparing the frame series settings:

WarpTools create_settings \
--folder_data frames \
--extension *.eer \
--folder_processing warp_frameseries \
--output warp_frameseries.settings \
--angpix 0.595 \
--exposure 3.5 \
--bin 1 \
--gain_path Ref/modified_8K-gain_ref.mrc \
--eer_ngroups 6

Because the Falcon4i does not generate 8K references (at least not to my knowledge), we are up sampling our gain reference in EMAN2 using e2proc2d.py. 

Using this method in the earlier versions of WarpTools (up to dev33) resulted in averaged mrcs after motion correction that were 4096x4096. It was my assumption that motion correction was happening at 8K, and then Fourier cropped to 4K.

I noticed this changed in the newer versions of WarpTools (v2.0.0dev36 and up)

I've read in the v2.0.0dev36 patch notes, that there was a bug with EER gain reference handling:

### Bug Fixes - **EER gain reference handling**: Fixed EER dimensions bug where the gain reference was not being properly scaled to match 4x super-resolution data. EER super-resolution is now fixed to 4x, and the gain reference is rescaled to match rather than relying on dimension-matching heuristics.

Is it now the case that all eer movies are downsampled to 4K before motion correction? Or is motion correction happening at 16K (matching the Falcon4i eer output), with automatic upsampling of the gain reference to match, followed by Fourier cropping back to 4K? 

It's not entirely clear to me how this process has changed in the newer version, so any insight from the community or Warp team would be helpful!

Cheers,

Robert