hydromodpy.core.rng.manager

Centralized RNG seed manager.

What

RngManager derives deterministic per-consumer child seeds from a single master_seed. Every code path that needs a random source (stochastic calibration, mesh point sampling, synthetic forcing, …) goes through this manager so that the same master_seed reproduces the same draws.

Why

Without a single entry point, each consumer instantiates its own np.random.default_rng(seed) from a config field that is not persisted in the catalog. A run is then unreproducible from the snapshot. The manager solves both problems: it is the only place where seeds are derived, and the master seed is persisted in runs_environment.rng_seed at registration time.