What the checker found
Red is wrong regardless of judgement. Amber is worth a look. Blue is a finding rather than an error — confirm it's real.
Section totals
Open each paper and type in the total printed at the head of the section. If the coded questions don't add up to it, a question was misread, missed or counted twice — and no amount of tag-checking will find that for you.
The questions
Work down the list. Tick each question when you've satisfied yourself the tagging is defensible — a question that mentions a concept isn't the same as one that assesses it. Questions marked Needs the paper can't be settled from what's in this file; the filter above gathers them if you'd rather do those in one pass with the paper open.
Send it back
This file contains everything you changed and why, anything still outstanding, and the corrected dataset. Email it back as it is — don't edit it by hand, since it's read by a machine at the other end.