Many good points have been made.
I would add:
- Get a good workstation. Laptop or stationary PC.
- Use calibre.
- Use epub as the main format. You can convert to any format and store all metadata with epub as the base format.
- Use ISBN to identify your books. And other widely used identifiers as available.
- Allow calibre and the OPFs to fully handle series and filenames and so on.
- Work locally, not over a LAN.
- Use a filesystem with real time bitrot protection. For instance ZFS or BTRFS with redundancy.
- Use good quality SSDs.
- Setup a good backup system. Snapshots. Multiple versions. Multiple locations. Multiple media. NAS, DAS and cloud as a minimum.
- Verify the backups regularly.
- Every year, make a full copy of everything using save to disk and catalog. Archive it using multiple media and multiple locations. Use rar-archives with error correction codes.
- Migrate to new hardware when the warranty runs out. Always! For EVERYTHING. This means that buying stuff with 5 years warranty or more is cheap and efficient.
For me this means Linux.
I don't have your requirements so I don't use ZFS or BTRFS with redundancy. I don't replace everything when the warranty runs out. And I don't make yearly archives. But other than that this is what I do. I only buy SSDs and HDDs that have 5 years warranty.
My current workstation for calibre, and more, is a Lenovo ThinkPad with 32GB RAM, a 1TB Samsung 970EVO Plus NVMe and a 2TB Crucial MX500 SATA for backups. Ubuntu Mate 9.10. I have all my finished calibre libraries on it. And incoming. Not junk and duplicates. With finished I mean that metadata is pretty normalized, the cover is nice and the book is in good shape. Can you ever fully normalize metadata for a calibre library with more than one book?
I run calibre using a script that automatically update a snapshot backup of my calibre libraries every time I run calibre. I update backup snapshots on a NAS, daily, when I charge.