I completely agree. That observation is actually what led us to build StateGuard. We expected model failures to dominate, but after reproducing a few dozen real

I completely agree. That observation is actually what led us to build StateGuard. We expected model failures to dominate, but after reproducing a few dozen real-world issues, most of them turned out to be contract mismatches between APIs, providers, schemas, tool outputs, and application logic. We're trying to understand which of those failures developers would actually want repaired automatically versus simply surfaced through observability.