A diff represents the changes between versions of files or other structured content. In repository analysis, it supplies code-level evidence that tests whether commit subjects and activity patterns reflect what actually changed. section 6.4 section 6.6
Use diffs to keep historical interpretation concrete:
- Pair commit metadata with selected diffs; subjects alone can mislead an analysis. section 6.4
- Sample landmark diffs across an era so the analysis checks its assumptions against source code. section 6.4
- Choose the detail level deliberately: metadata, paths, full diffs, additions, or deletions each expose different signals. section 6.6
Profile an era with landmark diffs
Represent each commit as month, author, scope, and subject, then attach five code-level anchors: the first commit, one from the peak of each third, and the last commit. Compare the model's inferred work categories with those diffs rather than accepting summaries based only on commit subjects. section 6.4



