Failed-write recovery lab
A test of what happens when a write succeeds but its response is lost.
Synthetic local lab, not a client deployment or a production model run.
The problem
Repeating a request after a connection failure can duplicate a change that already succeeded.
What I built
I built an approval and recovery path that keeps the same request identity after a restart. The test forces a lost response, then checks for the original result before repeating the write.
What you can inspect
The repository includes a container-based demo, failure tests, and verification notes.