What technique are you using to deduce what the content of the page is? A straight port of Arc90's open source code or some magic of your own? (I'm asking because I'm keen to improve their technique for a library I'm working on.)