Predicting whether released software will contain bugs or defects, or will not behave as expected is considered crucial in the software market, where the time to live of a release is considered to be around a year. It is reported that four weeks of delay in a release can cause up to 22 percent revenue loss. Bugs found in released software also affect the overall reputation of the project or even of the company. However, quality assurance in the days before the release is always problematic, due to constraints in implementing and testing the new functionalities. Having a model that, based on some metrics, can predict the bugs of the release will be crucial for assessing software quality.

The authors propose a model that uses behavioral metrics (for example, developer and team habits) that can result in software defects. For example, it is known in literature that frequent work interruptions and task switching affect software quality. The metrics are collected using Mylyn, an Eclipse integrated development environment (IDE) tool that enables collecting developer activities such as editing, invoking commands, selecting a file, and so on. The authors propose the evaluation of several metrics and conduct a study on an open-source project and in commercial companies.

The authors used data available from Mylyn tasks submitted to the Eclipse platform from 2005 to 2009 and four projects with teams spanning from one to 15 developers. On the data collected (which the authors acknowledge could be biased), statistical analyses are done, and the predictive power of individual metrics is extrapolated. One of the results is, for example, that the number of developers working on a single file and the number of edit events of a file were the most significant predictors.

The results of the paper are very encouraging (the micro-interactions metrics are promising defect predictors, and their solution worked well in both the commercial projects and in open source). A possible application proposed would be as an instrument for novice developers: they can learn what the bad habits are in their context by having assistance, for example, from Mylyn either online (under the form of a warning) or offline (as a report).