A variant I've been using for a while is leaving a failing test. When I get started again I can focus on getting that test to pass. In the process I build the bigger context.