I'm not sure it's a wrong question, or even a question of flaky code or buggy hardware, per se, but I do concede your following points. I could call autoflush or call flush manually.

While it's true that there is a degree of buffering by OS and even by DASD, you can still eliminate one degree of buffering if it's applicable. I'm used to writing system logs for audit and government organizations, so the accepted pattern is to flush often. I get that for this example it's probably over-kill. Especially given that the input file isn't destroyed/altered as it's processed.