Two of them you should manage carefully. One of them you should just remove. Most instructional design problems come from confusing which is which.
Intrinsic load comes from element interactivity — how many pieces of the task have to be held in mind and processed together, relative to what the learner already knows. Solving a quadratic equation has high intrinsic load for someone meeting variables for the first time, because every step depends on three or four other things at once. For an algebra teacher, the same problem has almost none — years of practice compressed those interacting elements into a single recognizable chunk.
This is the load you cannot design away. You can only manage it: sequencing simpler sub-skills first, segmenting a complex procedure into stages, or using worked examples before asking for independent practice. The content's difficulty doesn't disappear — it gets distributed across time and prior knowledge instead of being demanded all at once.
Same equation, same page, same intrinsic complexity on paper. The novice has to hold every step in working memory separately — the expert already has it compiled into one chunk. The expertise reversal effect: as knowledge grows, intrinsic load for the same material drops on its own.
Extraneous load has nothing to do with what's being taught — it comes from how it's presented. The classic case is the split-attention effect: a diagram on one side of the page, its explanation in a separate paragraph elsewhere, forcing the learner to hold each label in memory while their eyes travel back and forth to match it up. That search-and-match work consumes working memory capacity without teaching anything.
Unlike intrinsic load, this one is simply fixable. Integrate the label into the diagram instead of separating them. Cut the decorative chart that doesn't add information. Stop narrating text that's already on the slide (the redundancy effect). None of these changes make the content easier — they just stop charging the learner rent for navigating your layout.
Same diagram, same three labels. In split form, the legend sits off to the side and has to be matched by eye. In integrated form, each label sits at the part it names — the matching work disappears, and so does the extraneous load it cost.
Germane load is effortful processing aimed at building a schema — comparing two worked examples to notice the shared structure, explaining a step to yourself in your own words, predicting an outcome before checking it. It feels like work, and it should: this is the productive kind. The mistake is treating any difficulty as germane by default. Solving twelve near-identical problems in a row can feel effortful while building almost no new schema at all — that's just repetition, not integration.
Germane load isn't something you add on top. It's the capacity that becomes available once extraneous load has been removed and intrinsic load is properly sequenced — which is why the meter above shows it as the remainder, not a separate dial. Some versions of the theory since fold germane load conceptually back into intrinsic load itself, treating "productive effort" as one face of the same difficulty rather than a third ingredient — worth knowing if you see the two-factor version elsewhere.
The lever isn't making things easier. It's removing the difficulty that was never the content's to begin with.