A new method of detecting criticality from time-series data outperforms conventional metrics in the presence of variable noise levels for both simulated systems and real neural recordings.