This paper quantifies the spatial divergence between 128-channel Light Detection and Ranging (LiDAR) point clouds and Frequency Modulated Continuous Wave (FMCW) radar tracks in high-clutter urban environments using the TiAND dataset. Nearest-neighbor Euclidean distance between radar target centers and raw LiDAR geometry serves as the error metric, chosen because the dataset provides no semantic bounding-box annotations. Across all processed frames the system produced an RMSE of 10.083 m with a median error (P50) of 1.157 m, while the 99th-percentile (P99) deviation reached 43.008 m with the single worst-case ghost target exceeded 217 m. A total of 4,113 detections crossed the 15 m catastrophic threshold—a figure that must be interpreted against the full detection population reported in Section III. Critically, the top anomalies cluster across consecutive frames near fixed infrastructure suggesting persistent multi-path reflection geometry rather than isolated single-frame noise. These findings indicate that raw FMCW radar output without downstream filtering or LiDAR verification cannot be relied upon for spatial localization in unstructured urban traffic.