ScopeMap: An AI-Assisted, Human-in-the-Loop Workflow for Mapping Reaction Scope and Boundaries

Assessing the generality of synthetic methods is a cornerstone of organic chemistry, yet traditional manual selection and current optimization-driven algorithms often fail to delineate reaction boundaries, clustering instead in high-reactivity regions. Herein, we introduce ScopeMap, an iterative, human-in-the-loop workflow designed to efficiently map functional limits rather than merely maximizing performance. Leveraging a modified Centroidal Voronoi Tessellation (CVT) algorithm with a dynamic geometric repulsion potential, ScopeMap transforms negative experimental feedback into geometric constraints, actively steering sampling toward unexplored frontiers. Validated against a comprehensive dataset of biomimetic aldol reactions and a cobalt-catalyzed coupling system, the workflow achieves greater substrate diversity with a smaller selection of examples. By utilizing a representative subset comprising fewer than 4% of the substrate space, it successfully predicts the reactivity of the reaction space with an F1 score exceeding 90%. Furthermore, we establish the U-Score and R-Score-metrics derived from spatial entropy and mean squared distance (MSD)-to provide a standardized framework for quantifying sampling uniformity and representativeness. This work offers a resource-efficient paradigm for defining reaction generality, shifting the focus from exhaustive data enumeration to information-dense boundary mapping.