Left is not always image-left. If a person is facing the viewer, their left hand appears on the right side of the image. A text-to-image model can therefore render a plausible scene with the requested objects while still drawing the spatial relation from the wrong perspective.
FoR-T2I is the official benchmark release for Can Text-to-Image Models Draw from the Right Frame of Reference?. It… See the full description on the dataset page:
https://huggingface.co/datasets/ernie-research/FoR-T2I.