In my application, I have 3 models. Let’s call them X, Y and Z for
brevity. Their relationship are as follows:

One X has an arbitrary number (maybe zero) items of type Y.

One Y has 2 or 3 items of type Z. Moreover, these items are ordered,
i.e. I need to record somehow, which is the first one, which the second
etc. 
From a certain Z, I need to find out which Y it belongs to, and from a
Y, I need to find out which X it belongs to. 
When an X is deleted, the contained Y need to be destroyed too.

When an Y is deleted, the contained Z need to be destroyed too.
Here is what I have so far:
The mapping between X and Y is straightforward. When generating the
model for Y, I included a field X:references. Within X.db, I included a
has_many: :y, dependent: :destroy
For the mapping between Y and Z, I didn’t come up with a convincing
solution. When generating the model Z, I included a field Y:references,
but how do I model the 2 or 3 Z entities within Z?
From a database viewpoint, I would need three fields (with different
names, say z1, z2 and z3), which all contain the ID of a certain row
within z. The field z3 might be NULL (in SQL terms), while z1 and z2
must always be provided.
But how do I define this in the “ruby generate model” command? The only
field type which could store a reference to another table, would be
‘references’, but don’t see how this would help me in this case.
I already thought about this workaround: I do a has_many relationship
from Y to Z, and my application makes sure, that there are only 2 or 3 Z
elements assigned at one time. In addition, I add to my Z table a field,
which says whether it is number one, two or three within this group.
This should work, but can this really be called an elegant design?