Concerning historical domain data, there are several vendors (e.g. Domaintools) that provide this kind of information, and domain registrar information usually contains the registration date for the domain, so proving it was a different site is possible.

You're absolutely right though that it's highly problematic if we use automated tools to analyze this kind of data, as many people can end up in the wrong category due to either flawed analysis methods or faulty data.

Again I'd say that currently the greater threat comes from private companies that will use this data to build "shady" rating systems which will affect the daily lives of many people, without giving these people a chance to verify or oppose the data being used against them.

Never before I protected private companies, but looking at the history if Europe, I fear governments more.

Or what if someone, say on an innocent blog, places an invisible iframe pointing to a flagged website? Now every visitor has that website in their logs without ever willingly visiting the URL and seeing any of the content.

Which is why the measure used for criminal law is beyond reasonable doubt: if there is no other evidence than an historic visit to a domain that the police can't show contained criminal content then there's not going to be a conviction (indeed the CPS wouldn't even entertain carrying such a case). If on looking at your hard drive the police then find a cache of content supporting criminal activity you're certainly going to have a hard time if you're innocent.

Do you know of any UK caselaw covering situations where people were convicted on the basis of having visited a particular domain and no corroborating evidence was found? Would be interested in reading how that went down.

I was trying to point out that:

1. It wouldn't be hard to pollute the logs with noise, and

2. At the ISP level there's no distinction between "browsing a website" and accessing the URL unknowingly.

So let's assume no one will monitor these logs to fight thoughtcrime in real time, and it'll only be analyzed once someone becomes a suspect in a criminal investigation. Would it present as valuable evidence considering the points above?

As to your question: no I don't, but it isn't unheard of for law enforcement around the globe to set up honeypots. But then again there's no need to look into ISP logs because they control the server.