Abstract
Assessment and feedback are expected to achieve an expanding range of educational purposes. They must support learning, assure standards, foster learner agency, provide credible evidence of achievement, promote inclusion, prepare students for professional practice, and remain feasible within increasingly complex institutional environments. These legitimate priorities cannot all be maximised simultaneously, requiring educators to navigate trade-offs in assessment and feedback design. Generative artificial intelligence (GenAI) has brought these longstanding tensions into sharper focus by challenging conventional assumptions about evidence, authorship and academic integrity. However, the underlying design problem extends well beyond AI. This paper proposes the Dual Trilemma Framework as a conceptual lens for designing assessment and feedback for learning. Rather than offering another set of design principles, the framework positions trade-offs as the central challenge of educational design. The assessment trilemma foregrounds tensions between authenticity, observability and feasibility, while the feedback trilemma highlights competing demands of usability, sources and sustainability. Together, these trilemmas function as a meta-framework that complements existing assessment and feedback frameworks by supporting explicit, context-sensitive design decisions. The paper argues that improving assessment and feedback depends less on pursuing idealised design principles and more on making informed, transparent and pedagogically defensible decisions about competing educational priorities.