Simplify around the core question
Extra endpoints, rescue exceptions, or subgroup ambitions often dilute the study if they are not anchored to the primary decision being tested.
Treat heterogeneity as a design issue
Enrollment logic, phenotype assumptions, and concomitant care can all distort signal if they are left implicit.
Protect interpretation, not just statistics
A technically valid analysis can still leave stakeholders uncertain if the study was not designed with a clear interpretation framework in mind.