The Daily Project Details analysis provides a useful snapshot of project state for a single day, but does not provide much insight into trends during development. While we might notice when coverage suddenly drops from 100% to 50% in a single day, what if coverage has been slowly dropping over the course of six months?
Daily Project Details also does not give us much help in discovering co-variances among the various raw sensor data streams. For example, is it the case that high levels of simultaneous commits by developers tend to co-occur with build failures?
To provide the infrastructure necessary to investigate these dynamic, trend-based behaviors in development, we have implemented a "Telemetry" subsystem in Hackystat. The Telemetry system provides a language for defining ways to display sensor data and organize them into reports.
The screen image above shows a portion of one such telemetry analysis. The report provides three months of telemetry from the Hackystat software project, from August through October, 2004, on a weekly basis. This report produces eight charts, although only one is displayed in the above screen image. The displayed chart shows the number of successful builds and the number of failing builds per week. It reveals a period of volatility in development for a few weeks in September, when the number of failing builds spiked.