Sounds pretty cool although the project's homepage doesn't give a hint of a publicly available version: http://dub.washington.edu/projects/reform

The main innovations seem to be the example-based scraper and the focus on having end users decide where to apply the transformations. (What are some similar projects?)