This document discusses using sentiment analysis to predict project performance by analyzing language in project reports and communications. It proposes focusing the analysis on select correspondence between key project members, periodic structured reports containing issues/risks, and narrative management reports. Conducting a narrow sentiment analysis of reliable, high-confidence data sources from within the project domain can improve predictive capabilities over broad analyses by increasing the signal-to-noise ratio and computational efficiency. The meaning of words can depend on context, so sentiment analysis may need to consider the applicable contexts more narrowly when including a broader range of project text.