Very similar situation here. Months of agentic coding and seem to find problems in almost every test. And every model seems to find different failures. It’s a major issue if you don’t know what’s worth fixing and what isn’t.