Arena

Open a todolist with one entry per phase (Frame, Fan out, Cross-judge, Pick, Graft, Verify) before launching anything.

Phase A: Frame

Every candidate gets the same prompt, so the prompt is the contract.

1. State the artifact each candidate is producing.
2. Derive the rubric, 3-6 concrete gradeable criteria for what success looks like on this task. It is the picker's tool in Phase D. Candidates only see the task.
3. Pick the runners. Default to two, one each on your strongest judgment model and the strongest model of another family. If your harness offers one family, fill both seats with the strongest model you have and say so. Spawn more runners when the arena covers multiple design directions. Use the same model N times when the work is generation-bound rather than judgment-sensitive.
4. Give each candidate its own output location (a git worktree where possible, otherwise .scratchpad/arena-<slug>/candidate-<n>/), per the separate-before-serializing-shared-state principle (call the Skill tool with "principles" and read its reference).

Phase B: Fan out

Spawn all N subagents in one message, running in the background, each with the task, the path to the shared grounding, its own output path, and instructions to produce the artifact and a short rationale naming the alternatives it considered and what it rejected.

If a candidate produces no output, proceed with N-1 and note the dropout.