correct me if im wrong on this, but its possible that cutechess parses upperbound and lowerbound scores as the engines actual score.
ex. info depth 34 seldepth 49 multipv 1 score cp 26 lowerbound
in this case wouldnt the cp 26 be reported in the pgn as the actual score of the search? and wouldnt this affect adjudication as well? perhaps its better to ignore upperbound and lowerbound scores when parsing