Replication package for "Who Migrates the Schema? An Empirical Study of Database Changes in AI Coding Agent Pull Requests"

Replication package for the empirical study "Who Migrates the Schema? An Empirical Study of Database Changes in AI Coding Agent Pull Requests" (submitted to Empirical Software Engineering). The study analyses what AI coding agents (OpenAI Codex, GitHub Copilot, Devin, Cursor, Claude Code, Google Jules) do in the database layer (migrations, DDL, ORM schemas, raw SQL) of GitHub pull requests, and compares them with human pull requests from the same 889 database-bearing repositories and the same period (2024-12-24 to 2025-10-24). The package contains (i) code.zip: the frozen detector specification, mining code, human-baseline collector, independent re-implementation, statistical analysis (Python and R), revision analyses, all output tables and the human-validation labels; (ii) data.zip: sanitised per-PR and per-file derived tables sufficient to re-run the statistical analysis. GitHub person logins are pseudonymised (HMAC-SHA256 with a secret salt); bot and agent accounts are kept because they define the treatment. Patch text, PR titles/bodies and comments are not redistributed; scripts to re-collect them are included. Agent-side tables derive from the AIDev dataset (hao-li/AIDev, MSR '26; CC BY 4.0).