We used five Apache Commons projects as inputs for VTR, as follows.

• DbUtils
• Exec
• FileUpload
• Compress
• Collections

From the subjects, we collected the following test-case modifications for previously released source programs (in the HTML format).

• DbUtils
• Exec
• FileUpload
• Compress
• Collections

We then clustered them based on LCS-based code similarity. A figure below showed a dendrogram representing the clustering results. For short labels described in this figure, you can access their full information from our spreadsheet.

We then interpret these clusters (largest distance = 0.5 and minimum constituents = 2) and define the following "test-case modifications towards software release" patterns.

According these patterns, we implemented a validation functionality in VTR (see here).