In recent speedups, we migrated to using methods that operate on stacks of matrices (or list comprehensions where unavailable) to maximize performance. In doing so, we lost the ability to use different inverters for certain subsets of matrices that might be, e.g., singular or unstable for certain solvers. We might think about putting this functionality back in, though not at the expense of performance. Before investing the effort, we need to develop a set of test cases for solver instability.