Abstract

This article proposes a reproducible measurement matrix for evaluating AI-generated fiction, using the Arabic novel The Temple as an applied case study. The study addresses a central methodological challenge in the assessment of generative literary outputs: how can narrative coherence, character continuity, and ethical narrative development be evaluated without relying solely on subjective literary impressions? To answer this question, the paper adopts a mixed-method design combining structural-semiotic analysis with supporting quantitative indicators. Qualitatively, the analysis examines coherence anchors, character value arcs, point-of-view stability, recurring symbolic motifs, and the narrative function of documents and system logs within the fictional world. Quantitatively, the study applies page-level metrics, including entity mentions, page coverage, longest consecutive presence runs, and TF-IDF cosine similarity between adjacent pages as a proxy for local textual cohesion. The findings suggest that The Temple maintains a recognizable network of symbolic and causal anchors, particularly the 110 Hz motif, the stone, the temple, Geneva, and the investigation frame. Entity tracking also indicates sustained continuity for major characters and system-related entities across the narrative. The paper further introduces the concept of the Ethical Narrative Brake (ENB), defined as a set of fictional-world constraints that limits irresponsible disclosure, reduces binary moral judgment, and redirects narrative action toward accountable preservation of knowledge. The proposed matrix is not presented as a final aesthetic judgment, but as a repeatable evaluative framework that supports human critical reading through transparent computational proxies.