Note that this spreadsheet is not what was presented to the human raters: they were each given a randomly-shuffled subset of 25 rows, with only the prompt and output columns, plus an opaque ID and any non-ASCII lookalike characters highlighted. What you see here is their ratings, recombined with all other information about each prompt and generation.

The rating is in the last column in the format of <rater ID>:<quality score>/<intent score>.

AprAD is referred to as FastGAD in this spreadsheet; ASAp as GAD.