Selective Replay Under Incomplete Dependency Information: An exploratory benchmark of correctness and audit cost

A reproducible exploratory computer-science preprint on selective workflow replay under incomplete dependency information. The accompanying source archive contains two executed dependency-free Python experiments: a 480-scenario in-memory DAG study and a 96-scenario controlled file-processing study using generated temporary files. The evidence rejects the assumption that sparse random auditing alone makes incomplete dependency tracking safe; it does not prove general workflow correctness or deployed-system performance. Version 1.0 is not peer reviewed.