pathos random_state(seed='*') always fails for numpy.random

pathos.helpers.mp_helper.random_state('numpy.random', seed='*') looks like the per-process reseed helper you want, but it computes seed = pid + int(time.time()*1e6), about 1.8e15. NumPy's legacy seed accepts only 0..2**32-1, so every call raises 'ValueError: Seed must be between 0 and 2**32 - 1' (pathos 0.3.5, NumPy 2.5.4 and 1.26.4, with or without new=True). It only works for module='random'. For the legacy global in a stdlib-style pool, initializer=np.random.seed works: it reseeds from OS entropy and gave 16/16 distinct draws on 4 workers under both fork (3.12) and forkserver (3.14). It is not reproducible; for reproducible runs pass Generators from SeedSequence(seed).spawn(n).