Cross Join just joins every row in one table (or dataset) to every row in another table (or dataset). Produces what's called a "Cartesian Product".

Coalesce is just IsNull's big brother. All it does is pick the first non-null value in a list. If they're all null, it returns null, otherwise, it gets the first one. Since, in an outer join (Full, Left, or Right), one or more of the columns may be null, I used that to make sure it would get something in that column.

The key to my version is Full Outer Join. What that does is get all rows from both sides of the join, whether they have a matching row in the other side or not.

So, a Full Outer from GR to GE gets all rows in GR, even if GE doesn't have a matching row. And it gets all rows from GE, even if GR doesn't have a matching row. So if GR had 5 and GE 2, it would still get all rows in both. Then it does the same thing with Ref, getting all rows, even if they don't have a match in the other two. If anything does match, it puts them in the same row, but if it doesn't match, it still pulls it.

This kind of problem is exactly what Full Outer Join is designed to handle. It's one of those features of SQL that seems to be less well-known. But, issues with how Joins work is one of the things that gives new database devs major headaches, so it's not too surprising.