Datasets:
fix questions 519 and 520 (same window max and min width)
For the fzk_house the questions 519 and 520 had obviously wrong answers as was evident from the window EG-Fenster-6 being mentioned as having the maximum width aswell as the minimum width, which cant be true for a house with differently sized windows. The maximum is 2m and the minimum is 1m. Both OG Windows have a width of 1m, making them the only answer for question 520 for the windows with minimum width. The 9 EG Windows all have the same width of 2m, making them all equal contenders for the maximum sized window. So the answer for 519 has to be all 9 windows.
i uploaded the snapshot here so you can easily see the round OG windows and the EG windows of same proportions. To check the width you can search for IFC_WINDOW in the arc file at https://huggingface.co/datasets/sylvainHellin/ifc-bench/blob/main/projects/fzk_house/arc.ifc, and see that there are 11 windows, 2 of which are the OG round windows with a width of 1m and 9 of which are EG windows with a 2m width.
Verified against projects/fzk_house/arc.ifc: the file declares IFCSIUNIT(*,.LENGTHUNIT.,$,.METRE.) and contains 11 IFCWINDOW instances. EG-Fenster-1..9 are all (1.2, 2.) = 1200mm high x 2000mm wide, and OG-Fenster-1/2 are both (1., 1.) = 1000 x 1000mm. So 519 is indeed all nine EG windows and 520 is the two OG windows only. Your diff is exactly right and preserves the id column and the 1026-row count. Merged, thanks for the careful report.
For anyone using the dataset: this changes the expected answer for id 519 from 3 items to 9, so exact-match scores computed on that item before today are no longer comparable.