"We will not release any source code of the bot or the backend because we consider the risk of abuse as too high"
A great deal of raw data is available for download over BitTorrent. It shouldn't be too hard to test the validity of the census results by verifying a random sample of the results.
It's pretty rad. That's an effectively big botnet. I wonder if it's possible to have a breakdown of residential vs enterprise for the boxes he's connected to. Maybe a sample of the local subnet? Or doing port scanning on the internal network too, and judging based on network size?