Overview
Non-parametric tests are applied as a blanket solution to any data that "doesn't look normal" without considering whether non-normality actually violates the parametric test's validity (for large samples, the Central Limit Theorem makes t-tests robust to non-normality) or whether the non-parametric alternative answers the same question (Mann-Whitney U tests whether distributions differ, not specifically whether means differ). The decision to use non-parametric tests must be driven by the specific violation present and whether the non-parametric alternative's test question matches the research question.
The Non-Parametric Testing Framework applies the correct test based on design and violation type, reports rank-biserial correlation for effect size (the non-parametric analog of Cohen's d), and interprets results with precision about what the test actually tests.