Rank models with a blind LLM design-review council

Goal: Find out which model actually does my design reviews best.
Setup: agents used in source: Claude, Codex; tools/MCPs stated: orca.
Steps:
1. Build a council with an orchestrator (here orca’s built-in one).
2. Use a Sonnet 5 “president” agent to trigger members (Codex and Claude Code models) on the same review task.
3. Have members rate each other’s work blind.
4. Aggregate the votes; repeat across several runs before concluding.
Rules: ask me before anything irreversible (sending, posting, paying, deleting); don’t invent data, results or tools I haven’t given you; tell me what you couldn’t verify.
Based on: jinjingliang — https://x.com/JinjingLiang/status/2107537480945942954

Steps are from the source. Includes a “Based on” credit line.