The debate over whether a variable is best explained as a 'box' or a 'label' haspersisted in programming education for decades without resolution. We argue that thedebate is unresolvable in its current framing due to two conceptual confusions andmultiple uncontrolled methodological confounds. The first confusion is the failure tospecify language-semantic boundary conditions: the adequacy of any variablemetaphor depends on whether the target language uses value, reference, ownership,or immutable semantics. The second confusion is a category error that conflatesmetaphor (structural mapping) with causal model (mechanistic explanation of programexecution). Additionally, existing empirical work confounds metaphor effects withinstructional language, the mathematical interpretation of the = sign, and otheruncontrolled variables. We propose Six Diagnostic Questions as a framework forevaluating and designing metaphor research in programming education, and applythem retrospectively to the most influential existing study. The broader implication is areframing of variable pedagogy from 'which metaphor?' to 'how do learners constructcausal models of variable behavior in their specific language?'