v0.1.0

notearsforge v0.1.0 — world-class causal-discovery toolkit (author 晨星)

Headline (linear-Gaussian ER, sparse, er_p=1, 3 seeds):
GES skelF1 = 0.909 ± 0.074 vs PC 0.836 ± 0.051 -> GES significantly ahead (Δ=+0.073 > ½(σ_GES+σ_PC)).

Regimes (GES skelF1 ± std):
er/lin 0.909 | sf/lin 0.743 | er/nonlin 0.683 | sf/nonlin 0.638
PC comparators: er/lin 0.836 | sf/lin 0.554 | er/nonlin 0.699 | sf/nonlin 0.607.

18 pytest green · ruff 0.16.10 hard gate (lint+format zero-warn) · coverage 92% ·
determinism byte-identical · pinned deps (numpy 2.5.3 / scipy 1.18.1 / sklearn 1.9.1) ·
Docker build-time gate · CI matrix py3.12/3.13 × ubuntu/windows.

NOTEARS (ALM+L-BFGS-B) collapses to empty graph on the polynomial acyclicity
trap; GES is the production flagship. Scale-free hub & nonlinear SEMs are honest
stress tests, not headline wins.