A behavioral early-warning index for crowd-crush risk in confined pedestrian corridors: A simulation study of the 2022 Itaewon disaster

This is the authors' abstract. We don't add key points for this paper.

Crowd disasters are anticipated from density, which does not explain where or when control is lost. We reconstruct the 2022 Itaewon corridor at 1:1 scale and drive a multi-model agent simulation from measured transit and floating-population data, reproducing the recorded peak density (9.48 ± 0.59 agents/m 2 ) and the peak-loading time (22:17 against 22:15). We propose the Individual Control Index, which measures intent-based loss of maneuverability, and compare it with four established indicators across twenty-five simulations with operating points fixed out-of-sample. Pooled over the site, the proposed index ranks fourth of the five under both criteria, but pooling conceals a domain of validity: density- and pressure-based measures approach the ceiling in the open approaches (AUC 0.92) and degrade in the constricted corridor (0.74), where, at incident loading, three-quarters of cells are already dangerous, and detection ceases to be meaningful. Early warning, therefore, belongs among the approaches, where the intent-based index proved the most accurate under a mechanical danger criterion.