D-GAF: A Deployment-Grounded Socio-Technical Framework for Analyzing and Governing Real-World Generative AI Systems

Generative artificial intelligence (AI) systems have demonstrated remarkable capabilities under controlled laboratory and benchmark evaluation settings. However, real-world deployment consistently reveals a substantial gap between experimental success and sustained operational reliability. Challenges such as data distribution shift, adversarial user behavior, computational constraints, governance requirements, and continuous post-deployment adaptation often emerge only after deployment and remain insufficiently addressed by benchmark-centric evaluation.To address this gap, this paper introduces an experience-driven qualitative synthesis methodology for systematically analyzing real-world deployment from a socio-technical perspective. Drawing upon evidence synthesized from multiple representative deployment scenarios, the study identifies recurring deployment challenges, failure modes, and adaptation strategies through structured thematic coding, evidence traceability, and qualitative triangulation.The primary contribution is D-GAF (Deployment-Grounded Analysis Framework), a deployment-centered analytical framework that conceptualizes generative AI systems as evolving socio-technical systems rather than isolated predictive models. D-GAF is operationalized through a Deployment Matrix, a Deployment Readiness Checklist, and a structured deployment analysis procedure that support deployment diagnosis, operational monitoring, mitigation planning, and governance. The framework is further illustrated through a representative deployment scenario and supported by evidence traceability and a lightweight expert review.The proposed framework complements benchmark-oriented evaluation by providing a reusable analytical foundation for assessing deployment readiness, diagnosing operational risk, and supporting the trustworthy and sustainable deployment of generative AI systems in real-world environments. This work was conducted at Arab International University (AIU), Damascus, Syria.Official website: https://www.aiu.edu.sy