Then I proceeded backwards through the text, one line at a time. Backwards because it prevented me from reading the text, and missing valuable data. I would read a line of Q2 out loud, read a line from my transcript (out loud), and then re-read the line in Q2 (again, out loud). Then I would move on to the next line. Now maybe you can understand why this takes a while and you can't do it for very long successfully.
When I encountered a discrepancy between Q2 and my transcript, I would turn back to Q1. If my transcript matched Q1, I would make a note of the variation in blue pen on my transcript. If my transcript did not match Q1, I would note the variation between my transcript and Q1 in red pen, and then I would compare that with Q2, and if there was still a variation note that in blue pen. The double variant didn't happen especially often.
After I had made my way through the entire script, I went back to my digital copy and made my edits: correcting transcription errors from Q1, and making notes whenever there was a variation in Q2. On some very rare occasions I have chosen to modernize a word or use the Q2 variant, but in those cases, I have left a note in the text (using the notes feature, thank God for word processing) describing what Q1 actually says.
It is a long and tedious process. Now step back to 1608 where word processors don't exist, and type is set out manually from manuscript copy in a dirty, noisy environment where you're working on a deadline and get paid by the page. I made a lot of mistakes in my transcription of Q1, and thinking about all the tools I have available to use, as well as my comparatively voluptuously luxurious work environment, makes me a lot more forgiving of any mistakes that the original printers might have made along the way.