26 August 2026
The problem was never capacity. It was knowing which of the four drives on the shelf held the 2019 negatives, and whether that copy had ever been verified after the move.
What finally worked: a plain text manifest per drive, checksummed on write, and a single index
file that lists every manifest with the date it was last read end to end. No tooling beyond
sha256sum and a cron entry that shouts if a drive has gone more than a year unread.
Three drives failed the first audit. Two were bit rot in the same directory tree, which suggests the copy itself was bad rather than the medium.