Generalising HARF Beyond Bamiyan: A Bounded Cross-Site Evaluation of Generative-AI Reconstrucction at the Temple of Bel, Palmyra

The Heritage-Aligned Reconstruction Framework (HARF) was developed through the Western Buddha of Bamiyan to structure AI-mediated heritage reconstruction through evidentially constrained prompt design, prompt-sufficiency assessment and expert-in-the-loop evaluation. Its transferability beyond a figural-sculpture case has not been empirically examined. This article tests that question through a bounded cross-site evaluation of the Temple of Bel at Palmyra, substantially destroyed in 2015. The case is not treated as a second full reconstruction study, but as a test of whether HARF’s schema and evaluative logic can be instantiated for a monumental architectural site governed by different material, proportional and iconographic conventions. The Dynamic Prompt Blueprint was adapted by replacing Bamiyan-specific descriptors with Palmyra-specific architectural evidence, including limestone construction, fluted Corinthian columns, inscriptions, cella dimensions, pseudoperipteral colonnade, thalamoi and zodiac ceiling reliefs. A fourteen-item Palmyra block within the Phase I expert questionnaire was completed by 32 respondents. The findings show that HARF’s schema transfers without redesign, while dominant failure points shift towards architectural and decorative details, especially relief, surface, inscription and roofline form. HARF is therefore best understood as a diagnostic evaluation framework, not as a reconstruction engine.