From DevOps to XOps: Software and Reproducibility Artifacts for an Agent-Driven Reference Architecture
Authors/Creators
- 1. Fibabanka A.Ş
- 2. Süleyman Demirel Üniversitesi Mühendislik Fakültesi
Description
This record contains the software, configuration, generated datasets, raw execution logs, and analysis scripts supporting the manuscript “From DevOps to XOps: An Agent-Driven Reference Architecture for Autonomous Enterprise Operations”.
The package includes: (1) the Case Study 1 synthetic telemetry generator, Isolation Forest pipeline, parameterised fault campaign, LangGraph agent graph, prompt templates, 23 runbooks, 14 Open Policy Agent policies, infrastructure manifests, and the complete log and analysis code for 250 live executions of the reasoning graph against GPT-4o and a live policy engine; (2) the Case Study 2 NASA C-MAPSS FD001 preprocessing, LSTM training, simulated drift detection, automated retraining, promotion, and engine-disjoint evaluation pipeline, including results from five training seeds and 100 drift realisations; and (3) the Continuous-Time Markov Chain reliability model, escalation-aware analysis, sensitivity and robustness calculations, and indicative cost-analysis scripts.
The evaluation environment is controlled. Case Study 1 telemetry and fault scenarios are synthetic, and remediation actions are executed in dry-run mode without modifying a live Kubernetes cluster. Its behavioural metrics are measured from the 250 live reasoning-graph executions, whereas detection latency and recovery-time results are outputs of a parameterised simulation. Case Study 2 uses retrospectively labelled simulated drift and does not establish online production drift-detection performance. These artifacts therefore support reproducibility and feasibility assessment within the reported experimental scope rather than production-scale validation.
The archived raw execution log is the canonical source for the reported language-model behavioural results, because future executions against a hosted model alias may not reproduce generated outputs bit for bit. Reproduction instructions, data schemas, pinned dependencies, tests, and provenance information are provided in the README.
The original software is released under the MIT License. Third-party datasets, including NASA C-MAPSS FD001, retain their applicable source terms.