I think the current text is ambiguous for both ExtendOverwriteDefault and extend. How about something like this? (Note: properly rendering newline requires #1630)

 --metadata-delimiters METADATA_DELIMITERS [METADATA_DELIMITERS ...]
                         Delimiters to accept when reading a plain text metadata file. Only one delimiter will be inferred.
+                        Specified values will override the default list.
                         (default: (',', '\t'))
 --expected-date-formats EXPECTED_DATE_FORMATS [EXPECTED_DATE_FORMATS ...]
                         Expected date formats that are currently in the provided date fields, defined by standard format codes as listed at
                         https://docs.python.org/3/library/datetime.html#strftime-and-strptime-format-codes. If a date string matches multiple formats, it will be parsed as the
                         first matched format in the provided order.
+                        Specified values will extend the default list.
                         (default: ['%Y-%m-%d', '%Y-%m-XX', '%Y-XX-XX', 'XXXX-XX-XX'])

This makes me wonder why action="extend" is used over ExtendOverwriteDefault.

In all other usages besides --expected-date-formats, there is no default so it doesn't matter whether ExtendOverwriteDefault or extend is used and I think extend is used out of convenience.

For --expected-date-formats, the list ['%Y-%m-%d', '%Y-%m-XX', '%Y-XX-XX', 'XXXX-XX-XX'] might as well be hardcoded elsewhere in the code and the option be called something like --additional-expected-date-formats.

Not being able to override the defaults for --expected-date-formats seems like a bug, or at least an oversight.

I think the current text is ambiguous for both ExtendOverwriteDefault and extend. How about something like this?

That looks great! Let's do it. I don't think it's blocked on the newline formatting.

#1705 improves the help text for existing behavior. I still think we can consider replacing all of extend with ExtendOverwriteDefault, but maybe that should be a separate issue.