Louis? Well, seeing as this looks like your second G2G post, may I offer you a huge welcome! Anyone who's done any serious genetic genealogy in the past several years will know you, even if they don't immediately make the association. I sincerely hope you find time to drop by here from time to time.
And I agree with you. I wrote a script, as Steve evidently did, to strip out those three REFN fields. In all the exported GEDCOMs there is a top-level line under each 0 INDI, i.e.:
1 WWW https://www.WikiTree.com/wiki/Surname-123
Of course, that URL will import in various ways depending upon the application, but I can't see how having a separate REFN field for the WikiTree ID would be a problem.
Just FWIW, another matter eluded my ability to script for it. I did an quick experiment earlier this month and pulled my fairly small pedigree using the "limit for privacy" option. The export was 51,177 lines for 898 total records; 611 individual and 284 families. Running it through a GEDCOM validator threw a whopping 1,368 errors.
I doubt there's any feasible way WikiTree can mitigate that since all text in the biography box on profiles exports as one massive NOTE field. There were scads of non-UTF-8 compliant characters peppered in there, and many of those forced a line break which then also threw a second error because it broke the protocol of all lines under 1 NOTE needing to be prefaced by either 2 CONT or 2 CONC. I'm actually surprised that errors-to-lines ratios like that don't cause apps like FTM and sites like Ancestry to need the Heimlich maneuver when they try to eat the GEDCOMs.
Don't know how it might be accomplished, but stripping out those non-UTF-8 characters would be on my wish list for improvement of WikiTree GEDCOM exports.