Learning to Transfer School Attendance Zoning across Cities

TEXT START: Designing multi-school attendance zones requires selecting constraint settings that balance educational equity, accessibility, and school capacity.

The Dissection

This paper is an automation layer for repeated zoning calibration. COMPASS learns from prior search outputs, predicts zoning parameters for new cities, and hands them to a mixed-integer program that produces inspectable plans. Its real achievement is not educational reform. It is compressing a recurring expert-search task into a transferable prediction-and-repair pipeline.

The 99.80% solver-feasible rate and 49.14-fold deployment speedup measure computational throughput. They do not establish that the resulting zones are equitable, politically acceptable, durable, or educationally superior. The system accelerates the production of plans; it does not settle the conflicts those plans distribute.

The Core Fallacy

The paper risks treating parameter transfer and solver feasibility as proxies for policy success. They are not. A feasible zoning plan can still preserve segregation, impose unequal travel burdens, destabilize communities, or trigger political refusal.

COMPASS learns the compromises embedded in previous search procedures. It therefore automates inherited objectives and biases unless those are independently audited. It makes the search faster without proving that the search is aimed at the right target. This is the classic automation trap: mistaking a cheaper instrument for a solved social problem.

Hidden Assumptions

• Historical search-derived parameters are valid supervision rather than records of local compromises, modeling artifacts, or institutional bias.
• Urban representations contain the variables necessary to transfer policy across cities.
• A five-fold cross-validation design adequately represents genuinely unseen urban contexts.
• Solver feasibility is a meaningful success criterion rather than merely the absence of mathematical failure.
• Explicit parameters and spatial plans are sufficient for local inspection and adaptation.
• Enrollment balance can be treated as an optimization difficulty rather than a conflict involving housing segregation, capacity politics, transportation, and parental power.
• Conditional parameter repair preserves policy intent instead of silently changing the tradeoffs.
• Deployment-time speedup outweighs training, data preparation, validation, monitoring, and local governance costs.
• The stated objectives—equity, accessibility, and capacity—can be jointly optimized without a political authority deciding who absorbs the losses.

The abstract provides no evidence that these assumptions hold outside the evaluated benchmark or that affected communities accept the generated plans.

Social Function

This is primarily partial truth and transition management, with a layer of prestige signaling. It truthfully shows that AI can absorb a repetitive cognitive bottleneck and make institutional optimization dramatically cheaper. It also packages that displacement as a neutral technical improvement, leaving the contested distribution of benefits and burdens outside the model.

Under the Discontinuity Thesis, the important signal is not that COMPASS solves school zoning. It is that a specialized planning workflow—search, calibration, parameter selection, and part of the expert judgment surrounding them—can be converted into reusable machine inference. Human planners remain as objective-setters, auditors, political legitimizers, and exception handlers, but the economically valuable search labor is narrowed. That is a servitor role, not sovereign control.

The Verdict

COMPASS is a competent optimization accelerator, not a solution to educational inequality. It improves the speed of producing mathematically valid zoning plans while leaving the underlying scarcity, segregation, political conflict, and value judgments intact. Its strongest result is evidence for cognitive automation: once the objective and constraints are formalized, much of the recurring human search process becomes expendable. The machine makes the bureaucracy faster; it does not make the system just.