Well, this is a very good question—a very good extension to the original discussion.
First of all, I believe you do understand the original discussion correctly. I see no flaw in your analysis of the first and second version.
However, whereas in the original "Can am stand alone?" discussion we were looking at the conjunctions "and" and "so," we must now recognize that the conjunction "but" has the ability to throw a monkey wrench into the works.
Consider this "rule" from The Allyn & Bacon Handbook:
Use a comma (or commas) to set off a modifying element that ends or interrupts a sentence if the modifier establishes a qualification, contrast, or exception.
Now, the dependent clause in question, which includes a verb, is not strictly a "modifying element" as the AB Handbook conceives them, but there is a clear tendency across multiple contexts to set off contrasting elements.
So, the question is: does this dependent clause, which is clearly a contrasting element, deserve consideration as such, even though using a comma risks making the sentence appear as though the comma has been used incorrectly (or that the subject has been omitted incorrectly).
We find ourselves in the middle of gray area where using a comma would be appropriate and perfectly acceptable and where omitting it would be perfectly reasonable. However, both options risk being questioned in regard to their correctness.
Editors hate these situations. There are more of them than we probably care to admit.
In any case, the best option is to add the subject "I" to the second clause to make it an independent clause, requiring the comma (no matter what) and, thus, removing any cause for debate.
When dealing with artistic productions in which fidelity to "the voice" of the speaker is paramount, then the ambiguity of the situation must simply be accepted, and the editor must make a choice about whether to use the comma and then enforce that rule throughout the rest of the work. My guess is that there would be a slight majority in favor of using the comma in these instances.
Despite the fact this particular situation (contrasting dependent clause) is probably quite common, none of the references that I have appear to be brave enough to take on this very specific issue directly and precisely.