Sweep
The Lab drives one world down one path — that shows you what the policy did, never whether it is sound. A sweep runs a grid of parameter values across a fixed market and reports where the policy degenerates.
Every cell runs the identical workload, so a difference between cells is the parameter rather than noise; and every cell runs several seeds, so a verdict means the setting fails reliably. Degeneracy is judged relative to the market — m riding its ceiling under sustained inflow is the policy working.
steady_inflow — Consistent daily net buying — the case the policy exists to reward.
Pick a market and run the sweep. It executes off the main thread, so the page stays responsive while roughly four thousand simulated days run.