The paper introduces a new tolerance factor, $\tau$, which significantly improves the prediction of perovskite stability compared to the widely used Goldschmidt tolerance factor, $t$. $\tau$ is a one-dimensional, physically interpretable descriptor that correctly predicts 92% of compounds as perovskite or nonperovskite for an experimental dataset of 576 $ABX_3$ materials. The method, based on SISSO (sure independence screening and sparsifying operator), generalizes well to 1,034 experimentally realized single and double perovskites with 91% accuracy. $\tau$ provides a monotonic estimate of the probability of perovskite stability, allowing for the identification of 23,314 new double perovskites ranked by their probability of being stable. The work demonstrates the effectiveness of $\tau$ in guiding experimentalists and theorists towards the synthesis of new perovskites and highlights its potential for broader applications beyond perovskite stability predictions.The paper introduces a new tolerance factor, $\tau$, which significantly improves the prediction of perovskite stability compared to the widely used Goldschmidt tolerance factor, $t$. $\tau$ is a one-dimensional, physically interpretable descriptor that correctly predicts 92% of compounds as perovskite or nonperovskite for an experimental dataset of 576 $ABX_3$ materials. The method, based on SISSO (sure independence screening and sparsifying operator), generalizes well to 1,034 experimentally realized single and double perovskites with 91% accuracy. $\tau$ provides a monotonic estimate of the probability of perovskite stability, allowing for the identification of 23,314 new double perovskites ranked by their probability of being stable. The work demonstrates the effectiveness of $\tau$ in guiding experimentalists and theorists towards the synthesis of new perovskites and highlights its potential for broader applications beyond perovskite stability predictions.