A pilot on six questions and 117 passages.

  1. Ben, Nathan, Opus 5.5 (Medium) and GPT-6.1 Sol (Medium) each sort the passages for all six questions, without seeing each other’s work.
  2. Ben and Nathan compare their two lists and agree one list per question.
  3. Only then do the agents’ lists open. Ben and Nathan check how well each agent did.
  4. Ben and Nathan add any good positions the agents found. That gives the final keys.
Reader