Use-case page: RL environments and agent evals (SWE-bench)

[stubbi]

Context

Running thousands of isolated, resettable environments for RL rollouts and benchmarks is a proven, high-demand use case (Daytona, E2B, Runloop, Modal all market it). Fork-from-golden-state beats cold boot per episode, and ~3 MiB per fork makes the economics real.

Deliverable

• A page with the "spin N forks from one warm environment" hook and a SWE-bench style example link.
• The density and cost story (per-fork marginal memory) front and center.

Links

• Backed by mitos-run/mitos A3 (RL/SWE-bench example) and A5 (density benchmark).
• Part of the use-cases hub (C1).

Done when

• Page is live with the eval/RL hook and example link.