A piece of advice that is repeated here and elsewhere is to create atomic cards. I’ve long wondered whether that is good advice or not, or whether it depends on the subject domain.

The better I know a subject, the less atomic it feels. I don’t think and talk about the topic in discrete facts but in interconnected knowledge. This is why the complete independence between cards seems more like a weakness than a strength. The argument from the spaced repetition community is that purposeful linkages between cards builds in a contextual dependency. And that dependency makes the resulting memory contingent. But in some knowledge domains that seems more of a feature than a bug.