For SBE, we implemented a default limit of 10,000 reads for each candidate plan. This bounds the expense of the trial period cost at 10,000n where n is the number of plans – an important backstop given that we may incidentally trial the “wrong” plan(s) first. However, this limit can be much higher for the Classic multi-planner. As a result, the SBE multi-planner will sometimes do less partial execution than the Classic multi-planner, rendering it unable to distinguish amongst plans when no well-indexed plan exists.