The performance of commercial Li-ion batteries increases year after year. New mechanical designs of cells as well as changing electrolyte and active material compositions continually push performance boundaries forward. With such changes, the risks associated with thermal runaway of cells change too and a challenge persists in recording and benchmarking risks to determine whether increased performance comes at the expense of increased risk in the event of thermal runaway. The Battery Failure Databank provides a record of thermal and mass ejection behaviors of past and present commercial cells and provides a valuable resource for benchmarking behaviors and determining the predictability of risks associated with thermal runaway.A Fractional Thermal Runaway Calorimeter (FTRC), developed by the NASA Johnson Space Center in collaboration with the National Renewable Energy Laboratory (NREL), distinctly measures heat emitted from the body of commercial cells and heat that is ejected from the cell. The FTRC also facilitates in-situ high-speed radiography to link phenomena occurring inside the cell with external thermal measurements. The FTRC was used for all tests recorded in the Battery Failure Databank and high-speed radiographs accompany each thermal dataset. Having thermal, mass ejection, and internal dynamics behaviors from hundreds commercial cells under mechanical and thermal abuse conditions facilitates statistical analyses of cell behaviors and evaluation of the efficacy of machine learning methods for predicting risks associated with thermal runaway.Using data from the Battery Failure Databank, the following three topics will be covered in this talk: Correlations between heat generation and mass ejection: The total heat output from cells showed a strong correlation with the quantity of material ejected. An explanation for this based on distinct heat measurements from ejected and non-ejected materials will be discussed. Potential causes of outlier cells that produce exceptionally high heat: Upon conducting numerous repeat tests for each cell, outlier cells that produce exceptionally high quantities of heat were observed. High-speed radiography data revealed that internal dynamic phenomena can lead to extreme variations in thermal and ejection behavior observed externally. The predictability of thermal runaway outcomes using machine learning methods and limited data: With limited data on thermal behavior, mass ejection, and cell specifications, the accuracy of zero- and one-shot approaches to predicting the variation and distribution of heat output during thermal runaway of a new cell type, based on training data in the Battery Failure Databank, will be discussed. The analyses and open-access resources provided with this work are expected to provide a foundation for benchmarking the behavior and predictability of risks associated with thermal runaway of past and present commercial Li-ion cells. In addition to providing a valuable resource for designing safe battery systems, it is expected that this data will be imperative as we evaluate next generation cell designs and assess their respective failure mechanisms and associated risks.