Does LUKS encryption store data in such a way that any accidents involving loss of data in an unmounted partition would render the entire partition unusable, or is it possible to destroy just a few files atomically so it would still be possible to mount that partition and retrieve other files?

Backstory:

I was installing OpenBSD on a hard disk with about a dozen partitions: NTFS, FAT, EXT3, EXT4 and 3 partitions encrypted with LUKS. I made a mistake and told OpenBSD to use the whole disk instead of one partition. I realized my mistake some 4 or 5 seconds later, then I pressed Ctrl+c and killed the process. I suddenly had a blank HDD with a very large partition, a very small partition and no more Grub.

I had backups, but they were a few days old so I attempted recovery with a program called Testdisk. It worked. I recovered everything. I can even decrypt/open/mount my LUKS partitions.

Except that one of them took longer to comply. It refused to decrypt in the first couple of attempts. It finally gave in. Maybe I was mistyping the password. I don't know.

Anyway, I opened these partitions and tested many files. Everything seemed OK. After a couple of hours, I was editing text files in a particular directory with many text files and ran into one that is completely mangled, garbled, corrupt. Beyond any chance of recovery.

I have another copy of that file, but now I am apprehensive. Could any other file in those encrypted partitions have been corrupted by that accident? Or is that completely unrelated? I am wondering, does LUKS work in such a way that any corruption would destroy the entire partition, or is it possible to destroy just a few files atomically?

The way LUKS corruption works:

If a stream of random bites hits the filesystem when it is opened, the resulting corruption would be the same that you could expect from an usual filesystem.

If a stream of random bites hits the partition where the LUKS volume is placed, there can happen many things:

-- The LUKS header is affected. Unless you have a backup of the header, you have lost your data forever.

-- One or more blocks of encrypted data are affected. The result of this depends on the algorithm, initialization vector you use, etc. A 512 bytes block (for example) that has its twentieth byte corrupted is only readable up to byte 19, and the rest of it is lost. I think that, with most encryption configurations out there, the corruption would not spread to other blocks (Can someone confirm this?).

So, if corruption hit some data blocks, you could loose a few files, but hardly the whole filesystem.