Paper Presents RL Routing for Equitable NYC 311 Classification
Per the arXiv abstract (submitted 7 May 2026), Irene Aldridge and 26 coauthors present "Scaling the Queue," an econometrics paper that applies reinforcement learning to augment complaint intake capacity across six New York City Department of Buildings operational domains. The paper frames each domain as an MDP and trains agents that act as intake routers, assigning incoming complaints to action categories: escalate, batch, defer, inspect now. The authors state the system optimizes throughput, reduces misclassification cost, and explicitly includes equitable classification coverage as a reward objective. The paper reports post-hoc SHAP attribution showing complaint recurrence and neighborhood-level statistics are stronger predictors of actionable violations than raw complaint volume, and the authors note implications for routing given demographic correlates of those features.
What happened
Per the arXiv abstract (submitted 7 May 2026), the paper titled "Scaling the Queue: Reinforcement Learning for Equitable Call Classification Capacity in NYC Municipal Complaint Systems" by Irene Aldridge and 26 coauthors introduces an equity-centered reinforcement learning framework targeting intake classification bottlenecks in New York City municipal complaint workflows. The paper focuses on six New York City Department of Buildings (DOB) domains and describes agents that route incoming complaints into action categories: escalate, batch, defer, inspect now.
Technical details
Per the paper, each operational domain is formalized as an MDP where the objective function includes a term for equitable classification coverage alongside throughput and misclassification cost. The authors report using post-hoc SHAP attribution to analyse feature importance, finding that complaint recurrence and neighborhood-level statistics are stronger predictors of actionable violations than raw complaint volume.
Industry context
Editorial analysis: Municipal 311 and complaint-intake systems commonly face capacity-constrained triage, which can generate service disparities across neighborhoods. Research that treats equity as an explicit reward objective aligns with a recent wave of fairness-aware operational research and public-sector automation studies, where algorithmic routing is evaluated not just for efficiency but for distributional outcomes.
Context and significance
Editorial analysis: The paper sits at the intersection of applied reinforcement learning, public-administration operations, and fairness metrics. For practitioners, the finding that recurrence and neighborhood features out-predict volume has practical implications for feature engineering and risk modeling in intake systems. The work adds to efforts showing that model explainability tools like SHAP can surface operationally actionable predictors with equity implications.
What to watch
Editorial analysis: Observers should look for a full methods section and empirical results in the PDF (linked from the arXiv page) detailing training regime, reward weighting for equity, simulation or field evaluation, and any privacy or fairness audits. Also watch for replication on other municipal datasets and for discussion of deployment constraints in live 311 systems.
Domains covered (as listed in the abstract)
- •boiler safety
- •crane and derrick oversight
- •heat and hot water complaints
- •housing complaint triage
- •scaffold safety
- •Natural Area District (SNAD) protection
Scoring Rationale
The paper applies contemporary RL and explainability tools to a real-world municipal problem, offering actionable findings for practitioners in public-sector analytics. It is notable for combining operational MDPs with explicit equity objectives, but it is research-stage and focused on a specific domain, limiting immediate cross-domain impact.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
