generalfactneutralThe universal steepest-descent direction for covariance regret is the contrarian policy -(c-c̄)Machine Learning (Statistics)28 Jul 2026http://arxiv.org/abs/2607.18866v1