I've downloaded and run the program and can confirm that it lets me retrace my movements over the past 10 months with a scary level of precision. (Warden and Allan have obfuscated the data slightly in their application by showing information on a week-by-week basis, though they say that the data file it draws from goes down to the second.) The two scientists will be presenting their findings on Wednesday at the Where 2.0 conference in San Francisco.
According to Warden and Allan, the data seems to be compiled from cell tower triangulation, rather than GPS, which reduces its full precision (in my own investigation, I noted there also appears to be a log of Wi-Fi location data as well). My own map showed clusters around my home in Boston; the Macworld office in San Francisco, which I visit frequently; and other locations that I've visited in the past year, including Chicago and Houston. Zooming in further shows more detailed information, to the point of letting me isolate individual trips I've taken. And because the information is timestamped, you can theoretically even retrace steps on an extremely granular basis.
What technical reason justifies a 10 month track including WiFi spots? And cell tower triangulation can get you to within a couple hundred feet, not a few miles.
It reads the clock and puts a time stamp on the record. That's all. That's why it has an accurate time/date stamp on the file entry. Several people have examined the file and find ONE entry per cell tower: the latest time it was read. . . not every time it was read. Some claim that an error on their file handling (John Gruber) may have resulted in multiple entries. Most, have found only one entry per tower/WIFI site.