Designing school attendance zones means balancing equity, accessibility, and capacity, a process that usually requires expensive, city-specific tuning. A new arXiv preprint introduces COMPASS, a framework that learns from past zoning searches to generate usable parameters for unseen cities, reducing the need to repeat that calibration from scratch.
The system works in two stages: it first predicts interpretable zoning-control parameters from urban representations and difficulty signals, then feeds those parameters into a mixed-integer linear program that produces explicit spatial plans. If the initial solve fails, a conditional repair step adjusts the parameters. The authors evaluate across 3,000 US cities using five-fold cross-validation and a case study of Atlanta, reporting a 99.80% solver-feasible success rate and a 49.14-fold median speedup over city-specific search.
Not all outcomes transfer equally. The analysis shows that structurally difficult cities see the largest feasibility gains, while enrollment balance remains the hardest to recover. Because the method retains explicit controls and spatial plans, local authorities can inspect and adapt the output—an important property for a policy domain that affects families and communities directly. The authors frame the work as a step toward reducing repeated search while keeping human oversight in the loop.