HARKing

Exploratory analysis is legitimate and often valuable, so the problem is not looking at the data. The problem is the rebranding. When a pattern that emerged by chance is written up as the study's original prediction, readers lose the information they need to discount it. A confirmed a priori hypothesis carries far more evidential weight than a post hoc observation, and HARKing erases that distinction.

Reviewers can spot it when the introduction predicts, with uncanny precision, exactly the interaction or subgroup effect that the results happen to show, or when the framing is too clean given a messy design. A registration or protocol that names different primary hypotheses is decisive evidence. Absent that, an oddly specific prediction with no prior literature behind it should prompt questions.

The fix is honest labeling. Findings generated by exploration should be reported as exploratory and flagged as needing independent confirmation. Preregistration and registered reports make HARKing structurally difficult because the hypotheses are timestamped before analysis begins.