Wrap-up

	Frequentist	Bayesian
Returns	compatibility of data with a model (p, CI)	posterior over parameters; Bayes factor between models
Lets you say	“compatible with effects from A to B”; “rejects values outside”	“P(δ > 0) = …”; “the data favor theory 1 over theory 2 by k”
Needs from you	a planned procedure; a SESOI for null claims; exact p and intervals	a prior you can defend in print; a sensitivity analysis
Earns its keep in theory building when	you need long-run error control, falsification of values, equivalence tests	you compare specific theories, support boundary nulls, accumulate studies, model heterogeneity
Cannot fix	confounding, bad measurement, the wrong estimand	confounding, bad measurement, the wrong estimand

The frameworks disagree least about the estimate and most about what you may say. The design decides what either one can mean.