Sometime back, while trying to source and connected some isolated Slade profiles, I realised that they were part of a GEDCOM which had somehow gotten completely disconnected during the upload process. Then I learned that there are other GEDCOMs which have had the same thing happen, so I started the Lost and Found Project. Up until today, all the disconnected GEDCOMs that we have identified were from several years ago (2010-2012, mostly), so I had assumed that whatever was causing the GEDCOMs to be disconnected had been fixed, so that once we reconnected the existing disconnected GEDCOMs, there would be no more of them and we would be able to move on.
However, a new post caused me to realise that the problem has not, in fact, been fixed, so I suggested to the original poster that she add the "tech" tag to her post so Chris can see it, and hopefully troubleshoot why it is that GEDCOMs are (apparently still) getting disconnected on upload. I don't know whether the problem is due to some other site (or standalone software) not forming their GEDCOM files correctly, or whether the import tool isn't parsing the GEDCOMs correctly, or what. I haven't had this issue importing GEDCOMs from three or four different sources, but clearly other people have.
As of the last time I got a dump from Aleš's database for the Connectors Chat page, about 2.5 million profiles on WikiTree had no connections at all. That's about 5/6 of our total unconnected profiles, and nearly 1/6 of our total profiles. Granted, probably a number of those are people who sign up, decide that WikiTree isn't for them, and bail without doing anything. However, I'm guessing that a bunch of them are because people uploaded GEDCOMs that got disconnected, and they never managed to get them reconnected again. (Come to think of it, I wonder how many people have joined WikiTree, uploaded a GEDCOM, and then, when all the family members they uploaded never showed up in their family tree, assumed that WikiTree was useless and left, not realising what had happened to their upload.)
(There would also be a bunch of people with only one connection from those disconnected GEDCOMs, because for some reason, husband-wife connections somehow manage to survive the upload process. It's the parent-child connections which get broken during the upload process, and since siblings are connected through their parents, they get disconnected from each other, too.)
And, even though that is a discrete topic and may entail considerable work to fix, I can't help but add some more comments about GEDCOM handling that I've had simmering on the back of my brain for some time now:
When I uploaded my GEDCOMs, they didn't get completely disconnected like that, but when I learned about the Unconnected report, I discovered to my horror that a number of profiles (sometimes singles, sometimes small clusters) had gotten isolated because I skipped uploading profiles where the person already existed on WikiTree (as the GEDCOM import documentation recommends). Unfortunately, when I skipped importing those profiles, all their relationships got skipped, too, and thus I had to go back through my Unconnected report and reconnect those people manually.
Recently, somebody wrote to my wife because he was preparing to import a GEDCOM which includes some people who already have profiles on her watchlist. He suggested (and I agree) that he should import those people anyway (because his GEDCOM contains a lot of information that she doesn't have) and then merge the profiles.
I understand that merges are less than ideal, because too many redirects slow down the system. What I have long thought would be preferable would be a setup which, when a match is identified during the upload process, the information from the duplicate profile gets added to the existing profile, rather than either skipped or created as a new profile and then merged. (In a case where the existing profile isn't Open, then what I'd like to see is some kind of temporary holding record created, which would get added to that profile once a profile manager approves the addition.) Granted, there would still be editing cleanup to do, but at least no redirects, and the connections and data from the GEDCOM would be preserved.