Builds
A set for one job. The model row is computed from the scenario board at build time. Tool rows only reference items.
Build a Web frontend with this setModel, skill, components, icons, motion, and type in one set. The model comes from the LMArena WebDev board. An editor picked the rest by job. Each row can be swapped.Build an Android app with this setThe model comes from the current Android Bench version, 2.0 long-horizon. UI is Jetpack Compose. Repo rules live in AGENTS.md so the agent does not add more XML.Write CLI and scripts with this setThe model comes from Terminal-Bench 4.0. The rules go in AGENTS.md: a repeatable command first, then the script.Build a full-stack SaaS with this setThe model comes from Terminal-Bench on the CLI scenario, chosen by run cost. It is not a second scenario board. The Next.js, Supabase, Convex, and ReactBench evals stay separate and are not merged into one score.