Introduction

Urban residential systems involve tightly interrelated variables, including parcel morphology, energy consumption, facility distribution, and transit accessibility. When these indicators change together, outcomes may reflect structural interactions rather than simple one-to-one effects. This study proposes a structured framework that clarifies these relationships and limits interpretation to verified results.

Methods

A relational data model integrates parcel attributes, building energy intensity, public transit nodes, and multiple facility types. Network-based shortest-path distances measure accessibility. Parcels are linked in a knowledge graph when they share nearest-facility or network-path relationships. The graph functions as a structured evidence layer, enabling traceable queries for each planning question. The framework examines functional dependencies, compares high–low groups, and evaluates scaling patterns. All validated findings are stored as graph-based evidence. Under graph-grounded constraints, the LLM interprets results stepwise using only retrieved and verified evidence.

Results

To evaluate the framework, we apply it to two analytical questions in the Singapore public housing parcel.

First, although social services such as childcare achieve full coverage, road-network–based nearest-facility analysis reveals that many parcels share the same service node. By jointly examining building scale and the distance to alternative facilities, the framework identifies parcels with highly concentrated service dependency, locating areas of highest fragility beyond simple coverage metrics.

Second, transit accessibility appears positively associated with energy use. However, validation confirms that energy strictly follows Energy = 46.47 × GFA, meaning consumption is fully determined by gross floor area. High-transit parcels also have a higher average GFA, which explains the correlation. After controlling for GFA, the relationship disappears, revealing a density-mediated pathway (Transit → GFA → Energy) and rejecting a direct causal interpretation.

Conclusions

This framework integrates relational modeling, knowledge graphs, and graph-grounded LLM reasoning to provide a reproducible diagnostic method for parcel-scale urban planning.