StateProof CrashLab

Three bookings were expected. Three rows were stored. One customer still lost their booking. StateProof CrashLab is a tool for developers who need to know whether an AI agent can safely retry a write after a timeout. In the live demo, North’s first booking appears twice and South’s disappears. A test that counts rows passes. A test that checks who was booked and how many times catches both errors. You can run the same four writes against 32 database index designs in your browser. Counting rows lets eight faulty designs pass. Checking booking identities correctly classifies all 32. Across 174,080 SQLite outcomes, no shorter fixed sequence classified every design in this family. I also reproduced the missing booking on Redis Streams and NATS JetStream when the client used the wrong key. Developers can inspect the rows, download a failing Python regression test, repair the index, and rerun the fault. Try the experiment here: https://praneethhere.github.io/stateproof-meta-hackathon/intent-study.html