Deciding what not to keep

Keep what you cannot recreate and what you would miss. Everything downloadable again is a cache, not an archive, and treating the two the same is what turns 200 GB of real memories into 4 TB of noise.

Real but slow. A yearly checksum pass over a few terabytes catches it long before you notice visually, and the value is knowing which copy is the good one when two disagree. Without checksums, sync happily propagates the corrupted file over the intact one.