Bandit algorithms such as Thompson sampling (TS) have been put forth for decades as useful tools for conducting adaptively-randomised experiments. By skewing the allocation toward superior arms, they can substantially improve particular outcomes of …