Skip to content
Open access

Telemetry-Native Reliability Engineering for Agentic AI Systems: Topology-Aware Failure Correlation and Decision Traceability

Sep 2026 · International Journal for Sciences and Technology · Vol 17 · 0 citations · 18 references

TL;DR

This paper proposes Telemetry-Native Reliability with Decision Traceability (TNR-DT), a framework that combines OpenTelemetry-aligned signal normalization, topology-aware incident correlation, predictive change-risk scoring, deterministic policy routing, rollback-readiness checks, and decision-trace evidence generation.

Abstract

Agentic artificial intelligence systems combine large language models, tool-calling runtimes, retrieval services, application programming interfaces, policy guardrails, memory stores, and human approval paths. These systems create a reliability problem that differs from conventional microservices because failures may arise from model drift, tool misuse, dependency latency, invalid action plans, retrieval errors, policy denials, or release changes across the agent execution graph. This paper proposes Telemetry-Native Reliability with Decision Traceability (TNR-DT), a framework that combines OpenTelemetry-aligned signal normalization, topology-aware incident correlation, predictive change-risk scoring, deterministic policy routing, rollback-readiness checks, and decision-trace evidence generation. The study evaluates the framework through a benchmark-referenced reproducible simulation. It uses current public benchmark references, including ITBench, ITBench Trajectories, AIOps2025/RCA100, AgentOps-Bench, OpenTelemetry Demo Dataset, and GH Archive, to define operational scenarios, telemetry classes, fault categories, and change-event features. A fixed-seed AgentOps replay then evaluates threshold-local alerting, global time-window grouping, and TNR-DT across 100 seeds and 240 incident episodes per seed. TNR-DT reduced non-actionable alerts per incident from 8.69 to 3.11, reduced modeled mean time to restore from 101.81 to 54.23 minutes, improved top-three root-cause hit rate from 55.07% to 81.30%, increased evidence completeness from 53.00% to 96.08%, and reduced policy-exception escape rate from 11.30% to 2.57%. Change-risk prediction improved only modestly, with area under the curve increasing from 0.620 to 0.653. The stronger contribution is therefore not predictive superiority alone, but a traceable reliability operating model that links agent telemetry, dependency topology, policy decisions, rollback evidence, and incident reconstruction.

Read PDF

Similar papers

#explainable ai Book Open access Oct 2026

Poster: VeriRAN: Explainable and Runtime-Verified Multi-Agent Control for Trustworthy AI-RAN

VeriRAN is a lightweight runtime-verified AI-RAN control architecture that separates intelligence from authorization, and uses AI agents as intelligent action proposers that generate radio-aware and service-level-agreement (SLA)-aware decisions.

Osman Tugay Başaran, Falko Dressler · 0 citations
Open access Oct 2026

FROM AIOPS TO AGENTOPS: SERVICE-LEVEL OBJECTIVES AND ERROR-BUDGET-GOVERNED AUTONOMY FOR PRODUCTION AI AGENTS

Large language model (LLM) agents now run production workflows, including IT and network operations, yet they are usually monitored with signals designed for deterministic services: availability, latency, and error rate. An agent can satisfy all three while completing the wrong task. This paper extends AIOps-style serv...

A. Solanke · 0 citations
Conference Sep 2026

A cloud-native log analytics-based operating fault tracing method for power monitoring systems

Reliable automated control in cloud-native power monitoring requires faults to be traced before stale measurements, delayed alarms, or unsafe commands propagate through optical-fiber Ethernet links and service chains. This study addresses the research problem of locating the initiating field, communication, or software...

Peng-Yuan Wang, Yi-Min Wang, Xiao-Shuang Lin · 0 citations
Preprint Sep 2026

DIBench: Benchmarking Decision Integrity of GUI-based Mobile Agents Under Deceptive Injections

DIBench provides a unified, reproducible benchmark to quantify the risk of in-task goal deviation in mobile agents and enable comparable evaluations of safety defenses, as well as providing a decision integrity benchmark for measuring this risk in mobile agents.

Li Hu, Kang-Hua Mo, Ying-Zi Jin et al. · 0 citations
Book Open access Oct 2026

AgentChaosBench: Benchmarking External Fault Detection and Localization in Agentic Systems

AgentChaosBench is presented, a benchmark for detecting and localizing runtime faults in agentic systems from their execution telemetry, and its held-out labels and compact prediction format support reproducible comparison of LLM-based and non-LLM diagnosis methods.

Chen-Kai Zhang, Yi-Ran Li, Yi-Fang Tian et al. · 0 citations

A Safety-Bounded SDC-to-MCP Gateway for Medical AI Agents

An IEEE 11073 Service-Oriented Device Connectivity-to-MCP gateway that exposes metrics, alarms, context references, and semantic metadata as read-only resources, while representing selected action affordances as policy-validated dry-run tools is presented.

Bennet Gerlach, Stefan Fischer · 0 citations

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.