The paper provides an estimator at the algorithmic level that uses pooled isotonic fitting plus a per-problem correction term, and this estimator does not need to know τ, V or μ in advance, achieving the minimax rate as long as n≥2 and m≥3.