Keep moving while the work still remembers the last move.

There is a window after working on something where all the context is still loaded, the details are present, and you know where you were going. Returning inside that window is cheap. Returning after it has closed means rebuilding all of it before you can do any actual work, which is why long gaps cost far more than the calendar suggests.

With anything complex, aim for shorter gaps rather than longer sessions. Twenty minutes tomorrow beats three hours next month. The work remembers for a while, and that memory is worth using.