Supervised multi-agent discovery of high-entropy layered double hydroxides for selective ethanol-to-acetate electrooxidation

Autonomous materials discovery platforms must deliver not only efficient optimization across high-dimensional chemical spaces, but also trustworthy system design that ensures safety, traceability, and robust scientific decision-making. Here we present AIChemist-Alpha, a hierarchical supervised multi-agent framework for scalable autonomous materials discovery. The framework separates orchestration, validation, and execution into distinct functional layers and integrates three reproducible system capabilities: traceable decision chains, data-flood governance that converts raw electrochemical signals into validated decision-grade abstractions, and supervised safety gates for risk-tiered experimental actions. We apply AIChemist-Alpha to the discovery of high-entropy layered double hydroxide (HE-LDH) electrocatalysts for alkaline ethanol oxidation, a challenging optimization problem defined by a large combinatorial composition space. Through supervised closed-loop exploration, the system identifies Co0.35Fe0.23Mn0.05Ni0.32Zn0.05-LDH as an optimized catalyst. It delivers an operating potential of 1.3447 V versus reversible hydrogen electrode (RHE) at 10 mA cm-2 with acetate selectivity exceeding 97%, and furthermore the assembled device demonstrates a durability of over 300 h at 100 mA cm-2. These results demonstrate how trustworthy multiagent autonomy can accelerate catalyst discovery while maintaining scientific rigor, operational safety, and practical reproducibility.