<![CDATA[This study aims to describe the psychometric characteristics of the End-of-Semester Summative Assessment (ASAS) Mathematics items for Grade 3 elementary school based on three cognitive levels-knowledge and comprehension (L1), application (L2), and reasoning (L3) and to examine whether empirical item difficulty differs across these levels. Using a descriptive quantitative approach, responses from 81 third-grade students on 30 dichotomously scored items were analyzed with the Rasch model using Ministeps software. Results showed excellent item reliability (0.94; separation index 3.93) and adequate person reliability (0.78; separation index 1.86). Only one item (item 8) was identified as misfit, likely due to the dichotomization of a partial-credit complex multiple-choice item. The Wright Map indicated that the instrument was generally easy relative to students' ability, with item 17-classified as L2 (Application)-unexpectedly the most difficult item in the entire instrument, exceeding even the average difficulty of L3 items. Descriptive comparison showed that mean item difficulty followed the assumed order (L1≈L2<L3), yet within-level variation, especially in L2, was substantial, indicating that the a priori cognitive-level classification in the ASAS blueprint does not fully and consistently reflect students' actual empirical item difficulty. These findings underscore the importance of empirical verification through the Rasch model as part of the routine item-review process for elementary mathematics assessments.]]>
