Quantify label-propagation errors in AgentLoupe

Quantify the false-positive and false-negative rates of semantic label propagation in AgentLoupe, including over-tainting caused by dependency explosion and under-tainting caused by missed causal links, across diverse workloads with ground-truth causal chains.

Background

AgentLoupe propagates semantic causal labels across processes and persistent resources such as files and sockets to preserve associations across process-lineage, time, and session boundaries. The paper acknowledges that this mechanism is limited and that uncontrolled propagation can produce dependency explosion, while restrictive propagation can omit genuine causal relationships.

The authors identify the absence of quantified false-positive and false-negative rates as an unresolved issue. They specifically motivate experiments using ground-truth causal chains to measure over-tainting and under-tainting across diverse workloads.

References

Although this work has achieved a limited propagation mechanism and a user-customizable noise reduction mechanism, the risk of dependency explosion still exists, the false-positive and false-negative rates of label propagation remain unquantified.

— AgentLoupe: Kernel-Resident Semantic Causal Labels for AI Agent Observability  (2610.23233 - Bao et al., 16 Oct 2026) in Section 5, “Open Challenges and Conclusion,” subsection “Propagation fidelity”