Benchmarks

pre-registered comparisons of deep research apis. task sets, provider configurations and judging protocols are frozen and published before any run, and every raw run is public.