Rubriche IA-assistite per la valutazione formativa nella scuola secondaria di II grado: argomentazione di validità, feed-forward e audit di equità in chiave DD/UDL.

This paper presents an integrated model for the co-piloted use of AI in the design and implementation of rubric-based Assessment for Learning in a Grade 11 class in a technical upper secondary school. The model combines AI-assisted rubrics, differentiated in line with Differentiated Instruction and Universal Design for Learning (DI/UDL), an Advanced Assessment Moderation Protocol (PMVA) to support inter-rater agreement, and a descriptor-level equity audit based on extended DIF procedures. Within a mixed-methods, design-based research framework (Anderson, Shattuck, 2012), we collected indicators of efficiency (rubric prototyping time, feedback latency), measurement quality (criteria–level coherence, weighted κ) and equity among SEN subgroups (L2 learners, students with specific learning disorders, students with individualized education plans). Findings indicate reduced processing time, increased inter-rater reliability and narrower assessment gaps for student groups most exposed to learning barriers. The paper discusses conditions for the valid and equitable use of AI-assisted rubrics within an inclusive formative assessment framework.