You're not a fan of the old quirky update then? Much as I dislike the obtuseness of it, I like the performance and admire the ingenuity of it. It does seem to provide results orders of magnitude higher than other methods in specific circumstances - running totals say.
You got it!
When Jeff first published an article on that, many people (including me) reacted in the forums saying that this was not documented, not supported, and not guaranteed to return the correct results. He responded that as long as a certain set of prerequisites -I believe there were five or six in that list- was followed, results were always correct. This was then soon disproven by various people posting repro code where his method failed.
His reponse was an update to the article. Not saying that this method is dangerous and should not be used (well, I do believe that he included a disclaimer somewhere, but he still kept advocating the method), but by extending the list of prerequisites to disallow all the scenarios that were posted before. Unfortuntely, SQL Server version and build number and exact hardware configuration were still not included, even though they should be. At that point, I (and I think most others) more or less gave up - I could put in the effort to find another repro to disprove his claims, but I realized that he would then just extend the list of prerequisites.
I never saw the need for quirky update. I published a method that is almost as fast in most circumstances. I think it's somewhere in one of the enormous topics linked to Jeff's articles on this subject, but it's also published in the MVP Deep Dives book (the first one, not Volume 2).
That all being said - one other poster (I think it was Tom, but not sure anymore) found a smart way to extend Jeff's method to raise an error if the code would generate incorrect results. Again, buried somewhere in these forums. This is, of course, still firmly in undocumented and unguaranteed territory - but I looked very long and very hard at his trick, and I dare to say with 99.9999% certainty that his claim is reliable - code that employs Tom's trick will either populate correct results, or generate an error.