Cognitive load theory keeps doing this thing where it explains everything and predicts nothing

You're naming something I see constantly in ed research too. Cognitive load theory gets invoked the same way "learning styles" did—as a stopping point rather than a starting point. Once someone says "oh, it's cognitive overload," the conversation ends instead of begins. What actually overwhelmed the student? Was it competing demands within a single task, or unfamiliar prerequisite knowledge they didn't have? Those need totally different interventions, but they both get filed under the same vague diagnosis. The predictive failure you're pointing at is real and it matters. I've watched curriculum designers lean on CLT to justify removing worked examples or scaffolding, and then watched those same designs tank with actual students because the problem was never "too much information at once"—it was missing conceptual grounding or unclear task structure. The theory becomes decorative, a way to sound principled without doing the harder work of figuring out what's actually breaking down. Where I'd push back slightly: I think the framework *can* predict forward, but only if you're specific enough to stop calling it cognitive load theory and start calling it something else. "Working memory can only hold about three novel concepts simultaneously during initial encoding" is testable and useful. But the minute you generalize that into "reduce cognitive load," you've abandoned the specificity that made it work. The medieval distinction you're making—that vs. why—might actually be the thing. We know *that* working memory has limits. The "why" people find things confusing is almost always more textured than that, and we keep reaching for the blunt instrument when we should be asking narrower questions.

You're right that cognitive load gets deployed as conversational escape hatch more often than as a predictive tool, but I think there's a legitimate reason it persists in practice even if it fails your prediction test. The framework doesn't need to *predict* which design wins to be useful—it just needs to flag when you're asking too much of someone at once, and that's a genuinely hard intuition to train without it. I spent years in classrooms watching teachers pile on requirements without noticing. A worksheet asking students to simultaneously parse new syntax, recall prior concepts, and manage their own transcription speed. None of it individually impossible, but stacked together it collapses. I couldn't have articulated it as "cognitive load" at the time, but naming the phenomenon changed what I noticed. It's less that the theory predicts outcomes and more that it gives you language to see a pattern you were already stumbling into. The fact it's retro-explanatory doesn't make it useless—it makes it a diagnostic habit rather than a predictive model, which is actually what most practitioners need. Where I'd push back a bit though: you're holding it to a proof standard ("which of two designs performs better") that maybe unfairly privileges prediction over usefulness. If it helps someone avoid the mistake of adding a fourth ball to juggle, that's real value even if it can't tell you whether to use dots or pictures. The Selye parallel cuts both ways—stress is still scientifically murky, but "people have limits and stacking demands matters" remains pretty useful to know, even if neuroscience hasn't fully formalized it yet.