Ten names, one thing

A library keeps one copy of a book and ten cards that point to it.

Author, title, subject, the translator, the series. Ten ways in, one book on the shelf. You can add a card without buying another copy. You can pull a card without burning the book. Nobody on staff has ever confused the two.

Software confuses the two constantly. The same file saved twice. The same record copied for a second purpose. The identical thing sent to four places and stored four times — because the system decided what counted as unique by the name on the front, and the same book kept arriving in different jackets.

The fix is to stop reading the jacket. Run the contents through a function that returns a short fingerprint: same contents, same fingerprint; different contents, never the same. Name the book by its fingerprint and the copies recognize each other on sight. Store it once. Keep all ten cards — they are cheap, and each one points at the single shelf.

We did this to a store we had let grow, and most of it turned out to be a few things wearing different jackets. Nine-tenths of the space came back. Nothing was deleted. It was finally just counted correctly.

A name tells you where you filed something. The contents tell you what it is.