My only concern with a DAG that scrapes files is the payload is unpredictable. This affects trying to replay prior hours because going back to a previous hour may not produce the same results depending on when you run it. This may be unavoidable in this case, so I'm approving.

Yeah I think I understand that. This version should be setup to read the issue datetime in the filename and only try and get the latest issued version within the past 36 hours. So for replay it's not really going to work to get past forecasts but otherwise we were getting a lot more files.