I think there are two different answers to your question.
The first answer is that RM and much if not most other genealogy software does not actually connect parents and children together. Rather such genealogy software creates a sort of abstract and amorphous concept called a “family”. With the “family” in place, the parents are connected to one set of connection points in the “family” and the children are connected to the to another set of connection points in the “family”. What people in the real world would call a family would be the actual parents plus the actual children plus this amorphous and abstract data structure called a “family”. I think that genealogy software mostly tries to hide this complicated data structure from its users, but occasionally this complicated data structure does show through to the users’ detriment. You have experienced this problem yourself.
The second answer is that RM is trying to be user friendly by supporting the addition of one parent at a time in Pedigree View and in Family View. When the first parent is added, this abstract and amorphous “family” structure is created and only one of the connection points for parents is filled in. Then when you use Pedigree View or Family View to add the second parent on another one at a time basis, it’s very easy to do it in such a way that another whole abstract and amorphous “family” is created, this time with the second parent included and the first parent omitted. This is essentially what happened to you.
It’s not as user friendly, but if you connect parents and children using RM’s Add Parents tool or using RM’s Add Child tool, it’s much more likely that everything will be linked up correctly. Both parents will be joined up properly and at the same time with the abstract and amorphous “family” structure and all will be well.
This data model is called a lineage linked data model. I suspect it’s as old as is genealogy software. The model may even have been around before genealogy software. I don’t know for sure about that. On it’s face, the lineage linked data model seems to make sense. But to me, this data model actually includes an even worse problem than the one that you encountered. Namely, I don’t think that this data model provides a good place to attach evidence of parentage - for example, that Sarah Doe was the daughter of John Doe and Jane Smith. Let me give you a real example.
My paternal line great grandparents were Harley Bryan and Nannie Harrison. They had four children. The children were born at a time when official birth certificates were just becoming standard. The first two children did not have official birth certificates and the second two children did. For the third child Edna Bryan, I entered a citation for her birth certificate on the Parents line of her Edit Person screen in RM. For the fourth child Willie Bryan, I went into her Edit Person screen in RM to enter a citation for her birth certificate only to discover that the citation for Edna Bryan’s birth certificate was already there. And it was already there because Edna’s citation was not really for Edna. Rather, the citation was for this abstract and amorphous thing called a “family”. And when I added a citation for Willie’s birth certificate and then went back to Edna’s Edit Person screen, I could see the citations for both Edna and Willie’s birth certificate.
I’m just a simple country boy, but I think this data model is terrible. Edna’s birth certificate should be for Edna and Willie’s birth certificate should be for Willie. But it’s too late to change the data model and then to get the GEDCOM standard changed and and then to get all the genealogy software in the world changed to work with the new data model. Well, ancestry.com itself seems to be an exception in that it doesn’t seem to use the same model because it doesn’t seem to have this abstract and amorphous concept of a “family”. And the fact that it doesn’t seem to use the same model sometimes seems to wreak havoc with TreeShare when a Marriage fact is added to couple in RM and is not propagated properly into ancestry.com or vice versa.
Since I’m already on my soapbox, another place that I encountered this terrible data model was on the application form to become a member of the First Families of Tennessee. It’s like many other such organizations where membership is based on descent from some sort of distinguished ancestor or ancestors. In this case, the ancestors had to be already be living in Tennessee when Tennessee was admitted to the union in 1796. On the form, was a chart with two columns to be filled in. the first column was a list of names - the applicant’s lineage, beginning with the distinguished ancestor and ending with the applicant. Let me list just three of my generations.
Peter H. Bryan
The best evidence that Thomas Bryan and Peter H. Bryan were father and son is Thomas’s will which lists Peter as his son. The best evidence that Peter H. Bryan and Will Bryan were father and son is Will’s death certificate that lists Peter as his father. So I have three people and two documents, and I have to have have a document listed beside each name. As a math and computer nerd and somewhat of an expert in data structures, I think this is absurd. Nothing should be beside any of the names. Instead, Thomas’s will should be between Thomas and Peter. Similarly, Will’s death certificate should be between Peter and Will, viz.
will of Thomas Bryan
Peter H. Bryan
death certificate of Will Bryan
But that’s not the way the application form for First Families of Tennessee works, and to this day I don’t know how to fill out the form. To me, it’s the same problem as not having a proper place in RM or most any genealogy software to enter citations for the birth certificates of Edna Bryan and Willie Bryan to show that they were daughters of Harley Bryan and Nannie Harrison.