Today my heuristic is: the older the HTML the better. Often the horrendously crafted website were the work either of naive individuals that had no hidden agenda beside having fun with webpages, or passionated ones that would put long articles full of interesting facts.
Google's original algorithm of giving weight to inbound links was a very good way of surfacing good content programmatically but I don't think they ever solved how to make it work in an adversarial environment.
Google's other algorithmic approach is to measure user response to sites (what do they click on, how long do they view it, does it satisfy their query, etc). that algorithm is great for maximizing revenue but is bad at surfacing new original content.
In the end they also had to hand curate, which is what failed to work before them, and what they originally replaced, so really we are back to DMOZ and Yahoo now. Just with a very sophisticated monopolist as the gate keeper.
The hand curation is coming back and not only at Google and there's a weird feeling that in the end we want "humans" as much as "results". Even the most satisfying algorithmic result don't have the same as a collection made by a group (considering that algorithm cannot pass some SEO version of the Turing test of course).