In this paper, we use a developer-oriented statistical approach to understand what causes people in complex software development networks to decide to fork (break away), and what changes a community goes through in the run-up to a decision to break-up. Developing complex software systems is complex. Software developers interact. They may have the same or different goals, communication styles, or values. Interactions can be healthy or troubled. Troubled interactions cause troubled communities, that face failure. Some of these failures manifest themselves as a community split (known as forking). These failures affects many people; developers and users. Can we save troubled projects? We statistically model the longitudinal socio-grams of software developers and present early indicators and warning signs that can be used to predict an imminent break-up decision.