Judging between drafts
For anything beyond a straightforward request, Abbie produces more than one independent draft and an impartial pass judges between them, never a draft judging itself. You can control how many judges weigh in (one, three, or five) from the Model pool settings; more judges means more scrutiny, at a higher cost per request. If judges disagree and the answer fails a majority vote, it goes back for another attempt rather than being sent as-is. The best answer seen across every attempt is the one that’s kept, even if a later attempt is worse.Verifying the final answer
Separately from judging between drafts, a final verification pass checks the chosen answer against your stated preferences and the facts available before it’s delivered.When she tells you she isn’t confident
If, after all of this, the answer still scores low, Abbie doesn’t dress it up as more certain than it is. She says so plainly at the start of her reply, rather than presenting a weak answer with false confidence. This is a deliberate choice: an honest “I’m not sure about this” is more useful to you than a confident-sounding answer that turns out to be wrong.A run that takes noticeably longer than usual is often one that produced several drafts and went through more scrutiny before answering, not a sign that something’s stuck. See Watching her think to follow along live.