Zymera environment

The reusable simulator — a grid world of autonomous agents under partial observability, communicating over a limited-range, time-varying graph. It is deliberately independent of any learning pipeline: a policy is just a callable policy(obs, state, key) -> (action, state), so trainers can be swapped freely without touching the environment.

• worldgen · dynamics — terrain & spawns; movement, collisions, valid moves.
• comms — the comm graph (Chebyshev disk) + gossip channel; potential vs delivered edges.
• obs · sensor — partial-observability views and sensing footprints.
• missions · missions_terms — reward-as-data + the grouping / adversarial substrate.
• metrics — connectivity (λ₂), coverage, reachability, redundancy.
• env · rollout · viz — the step loop, episode rollouts, rendering.

Full API under Code reference. The learning / experiment code — which changes often — is documented separately and depends on this environment, never the other way around.