What happens when tool_error flags a tool?

A tool_error detection goes to triage first, which checks whether the rate shift is a real deviation rather than noise or something already explained.

If triage confirms it, the finding opens a case: one record per tool naming when the failure rate started shifting, which failure patterns moved it, and which call sites the traffic came through. That case is what a person is meant to act on, not the raw detection.

From there, once the project has a notification destination configured, the case reaches a human without anyone having to be watching a dashboard: a message with the case, what moved, and why it crossed the bar, sent to that destination. Nothing about a tool_error case is different from any other Tessary classifier’s case at that point; the classifier just supplies the finding.