Thirty Years Later, Boston's Algorithm Finally Got Fast

In 1999, Boston's school assignment system worked like a bad algorithm because it was a bad algorithm.

For thirty years, the district ranked student preferences and tried to spread demand evenly across buildings. But thousands of families got their last-choice school or nothing in walking distance, while other schools sat half-empty blocks away.

By the early 2000s, economists had proven that a different allocation system—the deferred acceptance algorithm—would match students to schools more fairly and reduce unhappy assignments dramatically. The problem was computational. Deferred acceptance required processing power school districts couldn't afford to run annually across thousands of families.

Those weren't abstract students—they were eight-year-olds assigned to schools their families had ranked fourth or fifth, or excluded from choice because slots had been allocated poorly. The harm was diffuse enough to feel natural, the kind of inefficiency cities accept. The system wasn't corrupt. It just methodically distributed poor outcomes because the mathematics available at the time couldn't do better at the speed required.

The system still won't ask them where they want to go.

This month, computer scientists announced a breakthrough in allocation algorithms that solves a thirty-year-old problem that became invisible because we'd accepted the technological constraint as permanent. The new method is provably faster than deferred acceptance while maintaining its fairness guarantees. Somewhere else, in a district still using the 1999 algorithm because faster solutions didn't exist until now, thousands of children are about to be matched to schools differently than they would have been yesterday. The system still won't ask them where they want to go. It will simply stop wasting their time on its own limitations.