Almost everything here started as an obstacle in my dissertation. A sensor that cost too much, a datalogger that did not log the right things, a forcing dataset that took a week to assemble by hand. Building the tool turned out to be the more interesting problem in most cases.