Task Misalignment in Expert Human Grading: A State-Based Case Analysis from IELTS Writing Assessment

This working paper examines a single-case discrepancy between expert human grading and a state-based inference system in IELTS Writing Task 2 assessment. Using a 48-state analytical framework, the paper demonstrates how high lexical sophistication can mask structural task misalignment, leading to systematic overestimation by human evaluators. Rather than proposing a new scoring model, the paper focuses on identifying a recurrent cognitive blind spot in expert judgement: the substitution of language maturity for task completion. The case is analysed as an illustrative and boundary example, highlighting where human intuition diverges from rule-governed task evaluation. This paper is intended to supplement ongoing research on state-based inference systems and is relevant to assessment design, AI-assisted feedback, and educational decision-risk modelling.