Indeed, you’ve found a bug! Thank you for reporting this.
Just to be clear, the intended behavior is for it to plot data at the finest granularity necessary. Basically this means that you’ll get a per-object plot unless both axes are of a higher granularity (ie: per-image or per-well).
The problem is that the function that checks what granularity of data is being plotted was checking the x-axis twice rather than checking both the x and y axes. As a result the scatterplot was returning image keys rather than object keys. So, if you select a single data point in your plot and select “show objects in selection” it SHOULD show you one object, but what it does instead (incorrectly) is show you all the objects that are in the same image as the selected object. This is the behavior that is supposed to happen when you are plotting per-image data because a single selected datapoint would represent a single image, so when you show the objects from that datapoint, you’d get all of them.
Fortunately there’s a workaround until a new release comes out: If you need to plot a per-image column vs a per-object column, make sure to put the per-object column on the x-axis.
Thanks again for your description of the problem, it made tracking this down very easy.