Back in the 2007-2008 timeframe when I added the streaming install/patch/play functionality to World of Warcraft, we also made it operate as an automatic repair

Back in the 2007-2008 timeframe when I added the streaming install/patch/play functionality to World of Warcraft, we also made it operate as an automatic repair system - if a disk block was read that had a hash mismatch, it just downloaded the correct block from the server (and wrote the fixed block to disk, of course). I looked at a month's worth of download logs, and estimated that 1 bit per GB per year flipped on hard disk data for our customers (we had already implemented re-reads before this point, because we had the hashes in place before we did streaming download, and so were pretty sure it was bad data on disk). Even at that fairly low rate, you can imagine that this reduced daily mysterious crashes by a nice amount. And keep in mind that hard disks already have error correction, so bits on magnetic media aren't as stable as you want. And, of course, this was not data at rest, there's lots of Wow play time going on, so lots of reads. I've been told that tape is more durable than spinning hard disks; for the sake of all those enterprise backups, I hope so. As a side effect, this made me increase how much redundancy I used for data I really cared about.