Automated essay scoring (AES) systems are often evaluated by how fast or how small they are, with compute cost treated as a purely technical efficiency issue. A new framework from arXiv challenges that assumption, arguing that cost should instead be considered a core part of measurement design.

The authors propose a "cost-aware measurement design" approach, suggesting that deployment decisions in AES need to go beyond reporting latency, model size, or inference performance. By reframing cost as a measurement concern, the framework aims to make cost a deliberate part of how scoring systems are built and chosen, rather than an afterthought.

The paper is available on arXiv under the title "From Efficiency to Sufficiency: The Cost-Aware Measurement Design Framework for Automated Essay Scoring." It does not present experimental results in the abstract, but instead offers a conceptual shift for how the field should think about resource use in scoring systems.