Researchers have proposed a decision layer for streaming systems that maintain a pool of expert models. The system decides whether to reuse an existing expert, spawn a new one, or defer based on statistical evidence. The authors present a method that makes all three outcomes statistically meaningful and prove its validity with finite-time anytime performance. They also propose a restarted e-detector that preserves lifetime anytime validity and controls multiplicity for unboundedly many experts. --- Why it matters: This research matters to AI engineers because it provides a more efficient and effective way to manage expert models in streaming systems, which is crucial for applications such as real-time prediction and decision-making. The proposed method can improve the accuracy and reliability of these systems by reducing false spawns and reuses. Source: https://arxiv.org/abs/2608.19888