I can all of R101 PNAS paper through the NUPACK script I posted last night and plotted the EV_classic_norm deltas for 1,2, and 3 kcal and then compared that to nupacks ensemble defect. I found a clear signal with the EV metric that was much more clear than the ED and much more useful. The EV metric filters out much more effectively as it filters out twice the amount of low performers. Here is a link to my plots. I can send you the script’ and libraries I used if needed.
Plots:
https://docs.google.com/document/d/1_jHJ4hS00qxGQZX4LJHSGx57VCgtfsYx4W-bd_sH9ig/edit?usp=sharing