Steel wheels mean a non-metal rail, for sound, it could work since we have light loads.
I mis-remembered the bearings being at “weird” angles. They are not all coplanar but they are not that weird. Maybe it was an earlier iteration that had more wild angles.
Theoretically this gets the magnet center to within 18mm (1.5 * pulley diameter) of the inner dimension of the box.
If the box is made of slabs of hardwood, say 3/4 inch (18mm) thick and maybe 3" wide, stood up on edge, then maybe you could cheat by hollowing out 1/2" (12mm) to give yourself more lateral room below the deck where the sand is. Then you can get the magnet center to within 18mm of the inner surface of the hollowed-out cavity, or 6mm from the inside dimension of the wooden box above the deck. If your ball is 1/2" or larger then you can touch the inner surface of the wood.
This is how the magnet center reaches 1.5 * idler diameter along the X axis:
And these are the rollers that would let it roll along the Y rails
At minimum Y position, the gantry idlers touch the corner idlers, so the minimum/maximum y position is 1.5 * idler diameter too.
The cable has no teeth so it has to be wound on a drum. That’s a separate conversation but in the grand scheme of things, there’s nothing difficult about that. (Yes I know there are cables with teeth but I don’t want to use them.)
What about LM8UU rails and bearings?
Isn’t that why we have gear sets?
Hard to joint rails and then we need two to make a gantry so it can’t freely rotate.
I wanted to give everyone a heart or emojie but I got cut off for the day. Been sitting in the hospital waiting for my wife to be done with Cataract surgery. So be active today
You liked 11 of my posts. Would you also like to buy me a drink? 
Wishing your wife all the best (and you too, obviously)!
^^^ what he said ^^^
How do I do that?
I was thinking about this… Do we really need 2?
What attaches the core… a rail to constrain it to the nominal X axis, and it has 4 belt end connections. Those belt connections have tension in order to constrain the movement to its position on the X axis, and the tension constrains the gantry to its position on the Y axis.
Obviously the belt attachment points must have a distance from the rail on at least the machine Y axis, possibly also on the Z axis, but not necessarily.
The belt tension ought to be enough to provide adequate stabilization to the core to keep the magnet in place. It might not be accurate enough if it were to hold a pen though.
I quite like the conduit Y rails with the belt inside it as an aesthetic, but the gantry can probably be further simplified.
Most bearings I have tried on a single round rail all have some inherent rotation to them. The magnet and ball might be enough with the belts as well. Higher tension though. I usually run my zen pretty loose to keep it dead silent.