A script becomes a tool when someone else can run it with their own inputs: python export.py orders.csv --limit 50 --dry-run. The standard library's argparse reads those words, converts them, and prints a usage message when they're wrong.
You describe the inputs once: a positional argument such as path, options such as --limit with a type and a default, and flags such as --dry-run with action="store_true". Dashes become underscores, so --dry-run is read as args.dry_run.
parse_args() reads the real command line. Here you pass a list instead, parser.parse_args(argv), which is also how you test a tool. When the input is wrong, argparse prints the problem and raises SystemExit. A good tool keeps its work in main(argv) and returns 0 for success and another number for failure.
Out: ('orders.csv', 50, False). The text "50" became a number, and the flag that wasn't given is False.