generalfactneutralCoupling in symbolic regression is not a general route to lower prediction error, but its main contribution is improved model interpretabilityNeural and Evolutionary Computing01 Aug 2026http://arxiv.org/abs/2607.26528v1