Rank items by pairwise comparison tournament

Rank large lists via pairwise tournament without loading all in one agent

Ranking hundreds of items by a nuanced criterion without exhausting a single agent's context window.

• · items to rank (array)
• · ranking criterion (qualitative description)

• · globally sorted list
• · comparison matrix (optional)

• · agent()
• · parallel()
• · schema + ranking logic

Workflow runtime; criterion unambiguous; items hashable (for dedup)

• · Bucket rankings disagree on order (ties)
• · Final merge drops items or introduces duplicates
• · Criterion too subjective, judges flip-flop

• · Pairwise tournament avoids token budget exhaustion
• · Bucket parallelization + final merge
• · Criterion stated explicitly (repeated to judges)