How can we figure out what is inside thousands of spreadsheets?

We have enough data today that we it may not be realistic to understand all of them. In hopes of vaguely understanding these data, I have been developing methods for exploring the contents of large collections of weakly structured spreadsheets. We can get some feel for the contents of these collections by assembling metadata about many spreadsheets and run otherwise typical analyses on the data-about-data; this gives us some understanding patterns in data publishing and a crude understanding of the contents. I have also developed spreadsheet-specific search tools that try to find related spreadsheets based on similarities in implicit schema. By running crude statistics across many disparate datasets, we can learn a lot about unweildy collections of poorly structured data.