Fetching the paper…

Kernel-based off-policy estimation without overlap: Instance optimality beyond semiparametric efficiency · Around