> But the most jarring issue for me is that it mangled all non-ascii filenames. They are stored as UTF-8 in the DB, but the client saves them as Windows-1252 or something. So I ended up with hundreds of gigabytes of files with names like фикац, and I can't just re-encode these names back, because some characters were dropped during the process.

> I wanted to write a script that forces Backblaze Client to redownload files, logs all files that can't be restored, fixes the broken names and splits restored files back into chunks to validate their checksums against the SQLite DB, but it was too big of a task for me, so I just procrastinated for 3 years, while keeping paying monthly Backblaze fees because it's sad to let go of my data.

Filenames are probably the most valuable of metadata for them to mangle. I value them as much as I do file creation/modification times. A backup program is dead to me if they mess up either of these.

I think it should be trivial for you to pipe your request into Claude now, and get them to write a quick script. Hope that'll free you from Backblaze for good!

  > I wonder if they fixed their client since then.

They have not. I spent more than a week trying to restore a little less than 2 TB backup because the client would just freeze at the last few % every time. I ended up having to break the restore into 200GB chunks on the web client and download and restore manually which was extremely frustrating and made me despise their (required) Windows client.

And they talked so much about how great redundancy they have on backend. I guess they don't count 404's