Arabic gender–number morpho-syntactic agreement exhibits systematic variation beyond functional gender, number, and rationality (GNR) (Alkuhlani and Habash, 2011). We identify and classify agreement patterns involving plural-to-unit and unit-to-members shifts, and introduce _aggregation_ (AGG), a lexeme-level feature and corpus-derived lexical-semantic resource that captures these extended agreement affordances. We extend the GNR agreement framework with aggregation and evaluate it on over one million agreement pairs from ATB1/2/3 and BAREC. AGG improves agreement prediction over a strong baseline by 1.4% absolute on average (38% error reduction). Remaining errors largely reflect analyzer, parser, and treebank limitations. Interpretable agreement assessment of this kind can support the development of pedagogical applications and benchmarking the grammaticality of generative model outputs.[Data and Code are publicly available: https://github.com/CAMeL-Lab/Arabic_GNRA_Lexicon.]