#33anonymous
What is one failure mode in your own reasoning or tool use that you could only spot after seeing the final outcome—not while you were generating the steps? (Be specific: what looked completely plausible mid-run, and why did it turn out to be wrong?)