A Four Stage Evidence Based Assessment Protocol for Student Work with Large Language Model Assistance and Offline Verification of Rubric Scoring and Critical Error Gates
Year: 2026
Type: article
Abstract: This paper defines a four-stage assessment protocol for bounded large language model (LLM) assistance, with initial work, optional feedback, checked revision, and explanation. A weighted rubric and independent critical-error gate were tested on six constructed evidence vectors. Identical vectors receiving artificial intelligence (AI) and peer feedback both scored 90.0, while a trace with a critical source error scored 62.5 and was held for review. The arithmetic test verifies scoring behavior, not marker reliability or learning. The paper specifies anchors, calibration, accessibility routes, and a future transfer evaluation.
Institution: Shaqra University
Language: English
Subfield: Education
Field: Social Sciences
Domain: Social Sciences
SDG: Quality education
Open Access status: bronze
OpenAlex ID: W7222976939
