RELIABILITY ENGINEERING WITH CAUSAL AND COUNTERFACTUAL REASONING FOR AI-NATIVE SYSTEMS
International Journal of Computer Science and Information Technology · 28.10.2026 · 18 (5) · 25-34
DOI: 10.5121/ijcsit.2026.18503 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Computer Science and Information Technology, online erschienen am 28.10.2026. Verfasst von Priya Ranjan Sahoo und 1 weiteren Autor:innen.
Abstract (Originalsprache)
Modern Site Reliability Engineering (SRE) relies heavily on observability systems that analyze logs, metrics, and traces through correlation. While effective for post-incident diagnosis, these approaches remain fundamentally reactive and insufficient for AI-native, machine-scale cloud environments. This paper argues that prevailing reliability practices are confined to the lowest rung of Judea Pearl’s Ladder of Causation i.e., Association and therefore cannot answer essential “why” and “what-if” questions. We present a formally defined causal reliability framework integrating structural causal models, graphtheoretic dependency representations, counterfactual reasoning, and AI-native reliability metrics directly into Service Level Objectives. We provide an original empirical evaluation on a 13-service microservice testbed using DoWhy’s GCM framework, comparing causal distributional change attribution against correlation-based and statistical baselines across six fault-injection scenarios. Our results demonstrate that while correlation-based methods achieve higher Top-1 accuracy (66.7%) in wellseparated fault scenarios, causal methods achieve 100% Top-5 accuracy and uniquely provide mechanismlevel attribution that distinguishes direct causes from propagated effects. Refutation tests confirm the robustness of causal attributions. We survey applications across e-commerce and finance, discuss statistical assumptions and sensitivity analysis, and propose operational frameworks for integrating causal inference into production reliability pipelines.
Zitieren
APA
Sahoo, P. R., & Prasad, S. (2026). RELIABILITY ENGINEERING WITH CAUSAL AND COUNTERFACTUAL REASONING FOR AI-NATIVE SYSTEMS. International Journal of Computer Science and Information Technology, 18(5), 25-34. https://doi.org/10.5121/ijcsit.2026.18503
Vancouver
Sahoo PR, Prasad S. RELIABILITY ENGINEERING WITH CAUSAL AND COUNTERFACTUAL REASONING FOR AI-NATIVE SYSTEMS. International Journal of Computer Science and Information Technology. 2026;18(5):25-34. doi:10.5121/ijcsit.2026.18503
BibTeX
@article{Sahoo2026,
author = {Sahoo, Priya Ranjan and Prasad, Sachin},
title = {RELIABILITY ENGINEERING WITH CAUSAL AND COUNTERFACTUAL REASONING FOR AI-NATIVE SYSTEMS},
journal = {International Journal of Computer Science and Information Technology},
year = {2026},
volume = {18},
number = {5},
pages = {25--34},
doi = {10.5121/ijcsit.2026.18503}
}
EUVIMED