Runs the A/B benchmark corpus and writes a results JSON.
mix bench.run --out results-head.json --ref general-updates
mix bench.run --out r.json --seeds 1,2,3 --budget-cap 120
mix bench.run --out r.json --only ok_small,rc208 --budget 2Options: --out Output JSON path (required) --ref Label for this run (default: current git branch) --seeds Comma-separated seeds (default: 1,2,3) --budget-cap Per-instance budget cap in seconds (default: 180) --budget Fixed per-instance budget in seconds (overrides size-scaling; useful for smoke runs) --only Comma-separated instance ids to restrict the corpus