Your agent harness is worth 15.36 accuracy points, more than the gap between some frontier models.

DataSpace benchmarks data agents on 410 cross-language tasks over 7,439 artifacts totaling 15.01GB across CSV, JSON, SQLite, Markdown, PDF and video, validated by 11 domain experts. Six frontier multimodal models across five frameworks: best accuracy only 66.34%, and harness choice alone produced a 15.36-point spread independent of model. The consistent failure across all six backbones was multimodal evidence integration and joins, which is precisely what every "chat with your data" product claims to do.