<![CDATA[Adaptive assessment systems depend on question banks whose difficulty rises in a credible way and whose items can distinguish learner performance. This study reports the design and validation of a 90-item question bank for adaptive learning in maritime mathematics at Politeknik Pelayaran Surabaya. The bank covered six STCW-aligned topics (speed-distance-time, ETA calculation, compass correction, great circle navigation, tidal height, altitude correction), arranged them across three progressive difficulty levels, and linked them to an AI-tutor system using a 2-up/2-down adaptive staircase algorithm. Validation had two parts. The maritime mathematics curriculum committee first conducted an expert pedagogical audit with a rubric covering curriculum alignment, difficulty progression, answer correctness, and linguistic clarity. Simulated pilot testing with synthetic learner profiles was then used to check whether the adaptive engine behaved as expected. In that audit, 15 of 90 items needed revision, mainly for clarity and difficulty calibration, while 75 were retained without modification. After revision, the bank still showed a balanced spread across topics and difficulty levels and maintained appropriate content validity. Simulation results likewise confirmed the intended pattern: difficulty shifted only after two consecutive correct or incorrect responses. For that reason, expert review paired with simulation offers a workable validation route for adaptive assessment systems in specialized vocational education contexts, especially when large-scale learner pilot testing is not yet feasible.]]>
