GeoContra: AI Framework for Verifiable Spatial Analysis
GeoContra is designed to ensure Python GIS workflows powered by LLMs meet geographic standards like coordinate semantics and topology. It treats each task as an executable geospatial contract, covering aspects such as natural-language questions, schemas, CRS metadata, expected outcomes, spatial predicates, and metrics. The generated programs undergo static rule checks, runtime validation, and semantic verification, with any issues entering a repair loop. In a study of 7,079 real geospatial tasks from 15 areas in Boston, across 9 task families and using 11 open-source models (with 600 runs each), GeoContra significantly improved spatial accuracy for closed models, boosting DeepSeek-V4's performance from 47.6% to 77.5% and from 57.7% to 81.5%.
Key facts
- GeoContra is a verification and repair framework for LLM-driven Python GIS workflows.
- It enforces coordinate semantics, topology, units, and geographic plausibility.
- Each task is represented as an executable geospatial contract.
- Contracts include natural-language questions, schemas, CRS metadata, expected outputs, spatial predicates, topology, metrics, required operations, and forbidden shortcuts.
- Generated programs undergo static rule inspection, runtime validation, and semantic verification.
- Violations are fed into a bounded repair loop.
- Evaluated on 7,079 real geospatial tasks across 15 Boston-area zones, 9 task families, and 11 open-source models.
- Improved spatial correctness on closed models from 47.6% to 77.5% for DeepSeek-V4 and from 57.7% to 81.5%.
Entities
Locations
- Boston