Cornell's Mixed-Integer-Programming Scheduler Eliminates Direct Exam Conflicts for 18,000 Students a Semester

For decades, Cornell University assigned final-exam slots using each course's regular lecture time, a method that produced hundreds of same-day and back-to-back conflicts every December as students juggled overlapping majors and minors.

Starting in Spring 2022, the university's Scheduling Team — led by operations-research professor David Shmoys together with Tinghan Ye, Adam Jovine, Willem van Osselaer and Qihan Zhu — replaced this with a mixed-integer-programming (MIP) optimizer that assigns each of roughly 540-600 final exams to one of 19-24 slots per semester, explicitly weighting course co-enrollment to avoid clustering students' exams.

Across five consecutive semesters through Spring 2024 (covering roughly 16,000-18,000 students each term), the optimized schedule eliminated direct same-slot conflicts entirely, down from 376-790 under the old lecture-time method, and cut 'triples' (three exams in 24 hours) by roughly 60-75%. The results were accepted for publication in the peer-reviewed INFORMS Journal on Applied Analytics, and the underlying scheduling complexity — over 2,500 total exams and deliverables across an 8-day period — is independently documented on Cornell's own Network Science course blog.

The approach is purely administrative: it does not touch instruction or curriculum, and the paper reports no measured effect on grades or learning outcomes, only on conflict counts and university operating costs.