Line-Level Post-Merge Code Retention Measurements for Merged GitHub Pull Requests - 356,650 merged pull requests in total

This replication package contains the intermediate and processed datasets produced for an empirical study of post-delivery code retention in pull-request-based software development. The study investigates whether the survival of code introduced by merged pull requests can serve as an early, repository-mined signal of delivery completeness and post-merge quality risk. The package includes line-level code birth data, retention measurements across post-merge observation windows, processing status records, and documented skipped cases. These artifacts support reproducibility of the code-retention measurement pipeline and enable independent verification, re-analysis, and extension of the study. The dataset was mined from public GitHub repositories using merged pull requests as the unit of analysis. Code retention is operationalized as the proportion of source lines introduced by a pull request that remain present after fixed post-merge windows. Processing status and skipped-case files are included to make coverage, exclusions, and measurement limitations explicit. We analyzed: 100 repositories in the discovered study sample. 91 repositories contained merged pull requests matching the study criteria. 356,650 merged pull requests in total. For the retention measurement stage: 214,198 pull requests had measurable newly added source-code lines and received retention measurements. 142,452 pull requests were skipped with documented reasons, such as no new lines in the analyzed source files, missing merge SHA, excluded/generated/large files, or other measurement constraints. Overall retention-processing coverage was 356,650 / 356,650 pull requests. Package contents line_births.csvLine-level records of source-code lines introduced by merged pull requests, including repository, PR identifier, merge commit, file path, line index, normalized line content, and test-path flag. retention.csvRetention measurements for PRs with measurable source-code additions. Each row reports how many introduced lines survived after a post-merge observation window, with retention ratios for 30, 90, and 180 days. retention_status.csvProcessing status for every analyzed merged pull request. It indicates whether the PR was successfully measured or skipped, together with counts of detected line births and retention windows. retention_skipped.csvDocumented skip records for PRs or files that could not be measured, including reasons such as missing repository data, missing merge commit, no source-code line births, missing snapshots, or excluded large files.