Iteration 02 of the GCM + O-information orthogonality-selection loop (S0–S5).
Thresholds are calibrated from a null on our own data, not inherited convention. Ground-truth effect-size separation: orthogonal ξmax 0.182 vs redundant ξmin 0.968 — the 0.80 cutoff sits in the empty middle (robust). The Type-I-calibrated cutoff is degenerate at grid n (significance cutoff ξ≈0.0039), so the gate is the separation. Sample perm-p: GCM 0.02, O-info 0.02.