We introduce a framework for evaluating a design against the data it could plausibly recover
The core idea is that the Best Plausible Dataset, or BPD, is the version of the study that applies the same research question and estimator to all *feasibly obtainable*, design-relevant data