What can no scheduling algorithm fix?

Badly written cards, cramming, and skipped days — and it can't turn memorisation into understanding.

• Card quality: spaced repetition needs atomic, well-formed items — one concept per card. No algorithm rescues a card asking you to recall five facts at once.
• Discipline: cramming everything before it's due, or skipping days so the review queue snowballs, breaks the model's timing assumptions.
• Scope: SR is for discrete facts and associations. It's no substitute for genuine understanding or for problem-solving practice — those need retrieval of a different kind.

Garbage in, garbage scheduled. The algorithm optimises when, not what.