Performance is not learning
This is the distinction the whole idea rests on. Performance is how well you're doing right now, during the study session. Learning is the durable change that shows up on a test weeks later. They are not the same thing, they are frequently *inversely* related, and you have direct access to only one of them.
So you optimise the one you can feel. You re-read, because your performance while re-reading is excellent — the words flow, nothing surprises you, you feel good. And you have optimised for exactly the wrong variable.
Difficulty is the signal
Retrieval that takes effort strengthens memory in proportion to that effort. Recall that is instant and effortless strengthens almost nothing — there was no work done, so there is nothing to consolidate. Which means the discomfort of struggling to remember is not a sign the technique is failing. It is the technique.
Hence: spacing (wait until it's hard), testing (produce it, don't recognise it), interleaving (never let the page tell you the method), blurting (no cues at all). Four techniques, one idea.
'Desirable' is doing real work in that phrase
Not all difficulty is desirable, and this is where people take the idea too far. Studying in a noisy room is difficult and undesirable. Illegible handwriting is difficult and undesirable. Material pitched far beyond your level is difficult and undesirable — you cannot retrieve what you never encoded.
The difficulty must be in the retrieval, not in the conditions. The test is simple: does the difficulty make you work harder at *producing the material from memory*? Then it's desirable. Does it just make everything worse? Then it's noise.