hard to say what is going on, you have a quite strange path, and most importantly I've seen in your GW input:
this is very dangerous because of the lack of completeness in the GW summation: this should start for band 1 and then bands should be added until convergence.
Code: Select all
% GbndRnge 15 | 25 | # [GW] G[W] bands range %
It is then possible that you have inconsistent corrections among bands providing inconsistent corrections among bands giving problems in the search of the Fermi level after the corrections are applied.
Anyway, without looking at the report I cannot say much as I do not know the entity of the corrections, the number of occupied states etc.