I mean, source code in files; how quaint, how seventies!
The community isn't nearly as afraid as it thinks it is.
If testing costs more than not testing, then don't test.
I tell people to start implementing when they are pretty sure there aren't more important stories out there. An iteration's worth of data is worth months of speculation.
Make it work, make it right, make it fast.
Learning research tells us that the time lag from experiment to feedback is critical.