The real question is why all but the last download do not have the correct checksum. It makes sense for it to delete downloads that do not pass that check. Have you tried checking for differences between the unsuccessful and successful downloads, e.g. different mirrors, file sizes or similar issues besides the checksum itself?
Maybe run something like Wireshark while you run that command to check for network errors and/or differences in the addresses?
Checking log files would also be an option to see if those contain additional information.
It might also be possible to inspect the different runs of the command via strace or bpftrace to get data like the file size or the checksum if it does not print those.