Interdependent constraints -- hall preferences, roommate requests, room attributes, adjoining suites, cancellations, turn-in order, multiple runs and upgrades -- makes automatic dorm room assignments a difficult global optimization problem. The deterministic approach apparently used by HMS suffers from myopia -- there may be thirty wonderful choices for the first student, but HMS's decision ignores the effect on the hundreth student who has fewer possible good alternatives, and so forth down the line. Simulated annealing's holistic approach seemed better-suited to the task of discovering an overall better set of assignments without hurting those who benefit from seniority. Initially, our goal was to see if that was true.