I haven't commented on it before, but I think your description of evidence based systems vs. conclusion based systems is the best description I've ever seen...
The "persona" terminology is new to me in the context of genealogy, and it seems very useful. I have wondered about the idea of storing what you are calling the "persona" data in a separate database where I would never do any merges at all. Instead, when I was ready I would drag and drop from the persona database into my main database where I would in fact do the merges. It just strikes me that after a while that managing the persona database (or persona databases if I had different persona databases for different sources) might become hard to manage. Nevertheless, the concept is intriguing and is definitely something worth thinking about.
Thanks for the kind words. The "persona" terminology goes back a ways. I'm not sure where it originated. The GenTech model used it. I think GEDCOM-X now uses it, though it might not be in the part of their model that they have developed.
Before there were genealogy programs, there was a fairly esoteric area of research, conducted mostly by demographers of various types, broadly called "nominal record linking." It still exists. In this context each "record" is information about a single person taken from a single mention of that person from a source, for example, birth, death or marriage records taken from a parish register. Think of a single index card for every mention of every person in every church register, city hall record, land record, tax record, etc, etc. Then think of encoding each of those cards into a separate entry in a computer file, and then think of algorithms to process that massive number of simple records, with the job of deciding which of those records belong to the same person and how those persons are related to one another. The purpose of the research might be to reconstruct families or to investigate migration patterns, all kinds of things. These algorithms seem to be commonly known as "nominal record linking" algorithms.
The term "nominal record" is used because the only thing each record has in common is having a name attached. All other data is specific to the type of source the record comes from. For me an evidence-based system would be a system based on these "nominal records" as their primary contents.
I think your idea of a second RootsMagic database to hold just your nominal records, aka personas, is fascinating. As you say, I think it would be hard to manage in the long term. It would consist of massive numbers of stand alone person records, and lots of little trees (child with its parents from birth records; parents with their children from census records; spouses and maybe their parents and maybe witnesses from marriage records; and so on).
Even with this idea you would have a bit of a problem when you wanted to "back out" a decision in your conclusion database. It might not be too hard though. You would have to remove all the facts from the person that came from specific nominal records.
But my laziness takes over once again. The RootsMagic approach of conclusion persons works very well most of the time. Backing out a persona from a person is not really much of an issue -- just edit or remove a fact or two from a person. I'm just too old to contemplate starting over.
One thing I have been fascinated with over the past years is the fact that there is such a well developed concept of the nominal record in the "professional" world of sociological research and demography, but the concept of the nominal record as the lowest common denominator of genealogical data, which it very much is, has never made it into the world view of actual genealogy programs. Persons in genealogical programs ALWAYS exist only at the conclusion level. I think the concept is just too much for the average hobbyist, and maybe even too much for most of the developers of genealogical software. (Or frankly, it's just NOT NEEDED.)
Gosh darn it, that's my grandmother. I know it's my grandmother. And you're telling me that you want me to create TEN DIFFERENT records for my grandmother just because she is mentioned in the ten different records I've collected about her. Come on man, you must be stark raving crazy.