This was an ... interesting question.
I don't call it a good question, because of the complexity - not the complexity of the subject matter, but that of the number of statements, and the number of rows. We had to deal with behaviour of a failed insert in a transaction, then the behaviour of two kinds of joins, union, ordering, and the position of NULL values in an order by (which, for the record, is not the same in all RDBMS's; whether NULLs go first or last is not defined in the ANSI standard, but left as an implementation-dependant choice). And all that on tables with three or four rows.
I think a good QotD should test one, maybe two subjects only. And preferably with a combination of statement complexity and number of rows that enables those with an understanding of the subject to work out the results in their head. This one severely pushed my ability to work it out in my head.
SanDroid, I'd love to see more questions from you, as the idea for this question is great - but as a suggestion for the next time, consider using two or three rows in the tables, not more. And don't add extra complexity by adding weird primary keys, failed inserts, etc. Just giving the table population, a query with a join on nullable columns, and some believable but incorrect answer options would have been enough.
As to the erroneous explanation - well, you already explained that this is the result of changes to question and explanation getting out of sync. Stuff like that happens, 'nuff said.
Hugo Kornelis, SQL Server MVP
Visit my SQL Server blog: http://sqlblog.com/blogs/hugo_kornelis