On the shelf of drives specifically: what you have is not a backup, it is a single copy of each job on a consumer drive with no verification. That is the thing to fix before deciding what to keep, because the current situation quietly loses work whether or not you have a policy.
The boring scheme that works: three copies, on two different kinds of media, one of them off site. For a one-person operation that is realistically a working drive, a second drive that mirrors it, and either a cloud archive tier or a drive that lives somewhere else. The off-site copy is the one that covers the things a shelf does not — theft, fire, water.
Two things people skip that matter more than the scheme:
Verify after copying. A checksum comparison, not a file count. Silent corruption on a copy is real and you find out about it when you need the file, which is the worst possible moment.
Spin the old drives up on a schedule. A drive that sits untouched for four years is a coin toss. Once or twice a year, plug each in, check it mounts, check a few files open. This is also how you discover which two are duplicates.
And index it. A plain text file or a spreadsheet listing drive name, client, date and job is enough — the evening you lost was an indexing problem, not a storage one.