Arena

Arena

Fan out N parallel attempts at the same task. Read every candidate end to end. Pick the strongest as the base. Graft the best ideas from the others into it. Verify the synthesized result.

Dispatch contract. Read provider-dispatch.md before fan-out. Configured values are provider-qualified descriptors, not host-native model slugs. The parent starts native and external lanes directly; children never route themselves. On Codex, resolve the remaining Claude tool names via codex-tools.md.

Start

Open a todolist with one entry per phase before launching anything. The arena runs autonomously and the list keeps phases from silently disappearing.

1. Frame
2. Fan out
3. Cross-judge
4. Pick
5. Graft
6. Verify

Phase A: Frame

The N candidates will receive the same prompt, so the prompt is the contract. Get it right before spawning anything.

1. State the artifact each candidate is producing.
2. Derive the rubric. State what success looks like for this task, then turn it into 3-6 concrete gradeable criteria. Concrete: Adds a --dry-run flag that skips writes. Vague: code is correct. The rubric is the picker's tool in Phase D; candidates only see the task.
3. Pick the runners. Use arena runners from the current harness's pstack model sheet when present. Otherwise default to claude:fable@max, codex:gpt-5.6-sol@max, grok:grok-4.6@xhigh, claude:opus@xhigh. Spawn more when the arena covers multiple design directions. Same descriptor N times when the work is generation-bound rather than judgment-sensitive.
4. Assign output paths. Each candidate writes to its own location (a git worktree where possible, otherwise /tmp/arena-<slug>/candidate-<n>/). N candidates writing to the same path is shared mutable state and fails the the separate-before-serializing-shared-state principle skill test.

Phase B: Fan out

Start all N lanes in one fan-out phase through the provider-dispatch contract. Native lanes are background subagents. External lanes are direct background launcher processes with retained task/session handles, never foreground calls and never subagents supervising subprocesses. Give every lane the task, the path to the shared grounding, its own output path, and instructions to produce both the artifact and a short rationale.

The rationale is mandatory. Without it, the parent cannot tell whether a candidate's structure is principled or accidental, which makes Phase E grafting unreliable. Each rationale names the alternatives the candidate considered and what it rejected.

An external lane counts only when its receipt says complete and carries either a matching provider-report or Codex's exact pinned-argv evidence; a native lane counts when its tool transcript returns the assigned model's result. If a candidate fails, proceed with N-1 and note the exact dropout in the synthesis record. Never replace it with another provider silently.

Phase C: Cross-judge

After all Phase B candidates complete, choose the judge descriptor from arena cross-judge pool in the current harness's pstack model sheet when present, otherwise from the runner defaults above. Prefer a provider different from the parent and the likely base candidate. Dispatch one read-only judge through the provider contract. It sees the rubric and completed candidates by path label, scores each criterion, and recommends a base with rationale. It runs in parallel with the parent's reading in Phase D, not with the candidates themselves. Starting it while candidates are still writing means the judge sees partial or empty outputs and reports them as dropouts.

Phase D: Pick a base

Read every candidate end to end before picking. Skimming N candidates surfaces only the candidate whose surface looks most familiar.

Score each candidate against the rubric criterion by criterion, not on holistic feel. Compare against the cross-judge. Agreement on the base confirms the pick. Disagreement means one of you is biased or the rubric was ambiguous. Read both rationales before deciding.

Pick the base on which candidate a future maintainer can extend most easily without breaking invariants. Prefer the cleaner boundary or smaller surface area when two feel tied, per the Laziness Protocol.

Record the pick and the reason in a short synthesis note alongside the base artifact, including the cross-judge's verdict.

Phase E: Graft

Walk each losing candidate once more and identify what is worth porting into the base. The signal is usually one or two things per candidate, not most of it.

Fold each graft in by hand, per the redesign-from-first-principles principle skill. Don't paste mechanically. The result has to remain coherent under one mental model.

Record what was grafted, from which candidate, and what was rejected and why. The rejection notes are the highest-signal part of the record. Future readers learn from what you considered and dropped, not just what you kept.

When N candidates converge on the same shape, that is a strong agreement signal. Note the convergence in the record and ship the consensus shape. No graft is needed. When N candidates wildly diverge, Phase A was under-specified. Reframe and re-run rather than averaging the divergence.

Phase F: Verify

The synthesized artifact has to hold up under the same scrutiny as any other output, per the prove-it-works principle skill. The arena does not earn you a pass.

If verification surfaces a problem the arena did not catch, either Phase A was wrong (re-frame and re-run) or one candidate caught it and you missed the graft (go back to Phase E). Don't paper over.

Outputs

One synthesized artifact. One short synthesis note alongside, naming the base, the grafts (with source candidate), the rejections, the dropouts if any, and the verification result.