The server code is on GitHub, but I don't know offhand where the offline geolocation scripts are. The Stumbler app's map will show your phone's GPS position (blue dots) and Mozilla Location Service's estimate (red dots) so you can compare the difference.
We store signal strength for possible future use, but our current location algorithms don't use it. Wireless signal strength is notoriously flaky. From some reports I've read, signal strength is more highly correlated with the user's orientation (i.e. is their body blocking the signal to the source) than with distance to the source.
I remember reading how Wigle Wardriving calculated fixes and the method seemed unsophisticated and lame.
For example, if I bike down a street then fixes would be detected 100m ahead of me and always pinned to the road at my current location.
If I bike down the road in the other direction the next time, will be fix become more accurate?
Normally higher SNR fixes should have more weight than weak fixes. Do they?