# MINI CHALLENGE: Fix the Broken Map – Rebuilding America's Homeless Resources Data

## Overview Across the United States, the infrastructure connecting homeless individuals to shelter, food, healthcare, and support services runs on broken information. Phone numbers lead to disconnected lines. Addresses point to closed facilities. Eligibility requirements listed online contradict what intake workers say in person. Service hours are wrong. Capacity numbers are outdated. This information crisis has real consequences. On a freezing January night in Chicago, a person experiencing homelessness might call five shelters listed as having beds available, only to discover all the information is months out of date. In Phoenix during a record heat wave, outreach workers waste hours driving to services that no longer exist at the listed addresses. The data problem literally costs lives. Multiple organizations maintain their own databases: city governments, the HUD Continuum of Care system, 211 information lines, nonprofit directories, and knowledge graphs built by research institutions. Each contains valuable information. Each also contains errors. They contradict each other. No single source is authoritative or current. This challenge asks you to fix the broken map. Take existing homeless resources datasets and knowledge graphs, identify what's wrong, and show how to make the information reliable. ## Background The homeless services ecosystem in America is fragmented by design. Federal funding flows through HUD to local Continuums of Care. State agencies run their own programs. Cities maintain separate systems. Nonprofits operate independently. Faith communities provide services without centralized coordination. This distributed model has strengths, but it creates an information nightmare. Researchers have built knowledge graphs (structured databases of relationships between entities) to map these resources, including the DREAM-KG initiative. These knowledge graphs can represent complex information: which shelters accept families with pets, which require sobriety, which have wheelchair access, which provide transportation to medical appointments. But knowledge graphs are only useful if the underlying data is accurate. The data quality problems fall into clear categories. Contact information changes when organizations move or update phone systems. Services get added or discontinued as funding shifts. Capacity numbers fluctuate daily but databases update monthly. Eligibility rules change but documentation lags. New organizations launch, existing ones close, and nobody updates all the different databases tracking this. Current solutions rely heavily on manual updates: someone calls each shelter monthly to verify information. This doesn't scale. It's expensive, slow, and still leaves data stale most of the time. We need approaches that use automation, cross-referencing between sources, crowdsourcing from service providers and users, and validation techniques that flag suspicious information for human review. ## What We're Looking For Your solution should address the core data quality challenge through one or more of these approaches: Data Audit and Error Detection: Take existing datasets (we can provide sample data or you can use publicly available sources like HUD's Housing Inventory Count, local 211 databases, or published knowledge graphs). Build tools to identify errors, inconsistencies between sources, outdated information, and missing critical fields. Show us what's broken and how you found it. Validation and Verification Systems: Design methods to validate information at scale. This might use automated calling systems to verify phone numbers, geocoding to confirm addresses, web scraping to check if organizations still operate, cross-referencing between multiple data sources to flag conflicts, or natural language processing to detect inconsistencies in service descriptions. Update and Maintenance Workflows: Propose systems to keep information current. Consider crowdsourcing from service providers, integration with existing case management systems, automated monitoring for changes, flagging systems that alert when data appears stale, or quality scoring that indicates confidence levels for different pieces of information. Knowledge Graph Enhancement: If you work with knowledge graph structures, show how to improve data quality within that framework. Build tools to validate relationships, detect logical inconsistencies (like a family shelter that doesn't accept children), merge information from multiple sources while preserving accuracy, or represent uncertainty and data freshness in the graph structure itself. You don't need to solve every problem. Focus on one aspect and do it well. A working prototype that validates phone numbers for 100 shelters is more valuable than a theoretical framework for fixing everything. Provide a demonstration using real or realistic data, documentation explaining your approach and how it could scale, and analysis showing what types of errors you found and how your solution addresses them. ## Expected Outcomes Successful solutions will demonstrate measurable improvements in data quality. Show us the error rate before and after your intervention. Quantify how many outdated records you identified, how many conflicts between sources you resolved, or how much more current your validated dataset is compared to the original. The ideal outcome is a pilot system that could be deployed by organizations managing homeless resources data: local governments updating their service directories, 211 information lines maintaining their databases, research institutions building knowledge graphs, or nonprofit networks coordinating regional information. Your solution should reduce the manual labor required to maintain accurate data, increase the reliability of information available to people seeking services, and provide transparency about data quality (helping users understand which information is verified and current versus uncertain or stale). Ultimately, we're looking for approaches that help ensure when someone experiencing homelessness needs shelter on a dangerous night, the information they find is accurate. That's the measure of success.