I rerender pages on demand like this, but also rerender the entire site every 6 hours. There's some computer science theorem that says tracking the dependency of which outputs are sensitive to which inputs is actually identical to solving the halting problem... i.e, no framework can actually track which pages are affected by which pieces of data. (This is true if you run custom code. If you are running in some type of limited non-Turing complete language it could be done.)