Zymera — learning pipeline & experiments

API documentation for the learning / experiment code — the parts that change as we iterate on training. Pulled straight from the source docstrings (Code reference in the nav).

• ctde_v0 — the MAPPO-CTDE training pipeline: policy/critic nets, PPO, connectivity glue (λ₂ metric + soft guardrail), config.
• t5lab — the distilled MVProp value-iteration planner, its arms (GPPN / Highway / MSP), and the distillation.
• planner_study — the validation / analysis harness (benchmarks, bracket, mechanistic proof, ablation).

The environment is documented separately

The reusable simulator — grid world, comms, dynamics, sensing — is the zymera package and lives in its own repo with its own docs, so it stays stable while this pipeline is rewritten: Zymera environment · code reference →