Authors

Kevin Lin, Zihang Jiang, Huayan Yu

Research Question

Can a property's documented historical provenance be systematically assembled, and is recognition of that history reflected in its value? The project builds and validates a reproducible data engine that turns property history into evidence, insight and a public address-lookup pilot for Baltimore's Greenmount corridor.

Background

Property-data systems describe parcels, assessments, permits and programs, but none systematically assemble documented history or test whether it affects value. Baltimore is running one of the nation's most ambitious vacant-property efforts (roughly 37,500 properties, with a $5B private investment target), and its development-reform agenda centers on making information centralized and accessible. No public tool yet does this for residents. The project is independent and is not commissioned by or delivered to the City.

The deeper stake is equity. Recent evidence (Patrick 2025) finds National Register recognition associated with a 13 to 14 percent price premium, suggesting documented history behaves like capital. Historically redlined neighborhoods were often denied both investment and documentation; on the Greenmount corridor, the administrative record undercounts vacancy by roughly 2.6 times, and history is undercounted far worse.

Methodology

The project involves:

• Working from one of Baltimore's best-documented commercial corridors: roughly 354 properties, 2,600 sourced archival incidents and a funded Maryland Historical Trust study producing per-parcel white papers.
• Building an ingestion and entity-resolution pipeline over deeds, newspapers, directories and registries.
• Constructing a temporal people-place-event knowledge graph.
• Developing a pre-specified, auditable Provenance Evidence Score (verifiability, density, salience, distinctiveness, connectivity), coded blind to prices.
• Validating the engine with spatially blocked held-out testing against conventional property-value baselines (out-of-sample R², MAE, calibration) and a connectivity-versus-proximity test.
• Applying a pre-committed decision rule: a monetary interpretation is shown only if validation thresholds are met; otherwise the pilot ships the evidence layer alone.
• Conditional extensions, only if feasibility gates clear: a National Register capitalization analysis, equity differences across historically redlined areas, and an improvement-scenario model.

Deliverables

• Algorithm development and design
• GitHub repository with documentation
• Interactive data dashboard
• Academic paper
• Data visualizations
• Policy memo
• Machine learning model

Data Sources

• Deeds, newspapers, directories and registries
• About 2,600 sourced archival incidents for roughly 354 Greenmount corridor properties
• Maryland Historical Trust per-parcel white papers
• Conventional property-value data (for baselines)