BEACON: 430 GB Valorant dataset stress-tests behavioral authentication under esports load
Researchers released a 430 GB multimodal dataset synchronizing mouse, keystroke, network, and screen data from 79 competitive Valorant sessions to benchmark continuous authentication under high cognitive load.
Continuous authentication systems need stress tests that go beyond typing patterns on static forms—and a new dataset built from competitive first-person shooter gameplay may provide exactly that proving ground.
BEACON (Behavioral Engine for Authentication & Continuous Monitoring), released this week on Hugging Face and GitHub, is a 430 GB multimodal dataset capturing 102.51 hours of active Valorant gameplay from 28 players across 79 sessions. The dataset synchronizes high-frequency mouse dynamics, keystroke events, network packet captures, screen recordings, hardware metadata, and in-game configuration context—all collected under the motor precision and cognitive load inherent to tactical shooters. The full on-disk footprint reaches 461 GB when auxiliary Valorant configuration files are included.
The dataset spans players of varying skill tiers, a design choice meant to surface how behavioral signatures shift with proficiency and playstyle. Valorant's demand for sub-100ms reaction windows and continuous spatial reasoning creates what the authors describe as a "rigorous stress test" for behavioral biometrics—conditions far removed from the controlled typing tasks that dominate existing benchmarks.
The release targets research in continuous authentication, behavioral profiling, user drift over time, and multimodal representation learning in high-fidelity esports settings. The full dataset, code, and documentation are available on Hugging Face and GitHub. Authors include Ishpuneet Singh, Gursmeep Kaur, Uday Pratap Singh Atwal, Guramrit Singh, Gurjot Singh, and Maninder Singh.