The open‑source Tycoon Learning Environment (TycoonLE) provides a reinforcement‑learning sandbox for long‑horizon logistics planning. Agents allocate capital, construct routes, move cargo, and manage debt within a fixed‑shape interface that supports JAX transformations like jit, vmap, and scan. A built‑in replay UI visualizes route choices, financing behavior, and profit trajectories, enabling deep audit of policy decisions.

Developers install TycoonLE with Python 3.11/3.12 and launch a quickstart that resets the environment, selects the highest‑scoring legal action, and steps forward. The repository ships a companion benchmark suite, TycoonBench, which reports agent performance across planning tasks. An example PPO training run runs in seconds on a few environments, demonstrating the framework’s efficiency for rapid prototyping.

TycoonLE draws sprite assets from the OpenGFX set used by OpenTTD, linking classic transport simulation aesthetics with modern differentiable programming. By exposing financing timing and delayed rewards, the platform targets research on action legality and audit‑ready RL. Researchers can cite the software via its provided BibTeX entry, ensuring proper attribution for future academic work.