Every FrontierSWE V2 run uses the same harness: the same shell and submission tools, the same 20-hour wall clock, and the same isolated environment per task. Scores compare models, not scaffolding.
Full documentation coming soon.