Currently when sources change all tests are rerun. It should be possible to track which tests depend on which sources. If a source changes, determine which tests depended on that source. Rerunning just those tests should make for a better --watch experience.