OptaPlanner Bin Packing with Three Planning Entities: Is my Model an Efficient Modeling Approach? - modeling

Question

I am concerned I am not modeling my problem well. I think I may be able to get away without using shadow variables and that I should be able to have aliquots and boxes as the only planning entities. Could you all suggest changes to improve my approach?
I am optimizing the assignment of sample aliquots into boxes and then these boxes into the racks of available freezers. There are hundreds of thousands of aliquots, hundreds of boxes, and tens of freezers.
More aliquots are collected over time and the number of boxes used to store these aliquots cannot be known ahead of time because there are two vial heights, each requiring a different height box. The taller box takes up the room of two shorter boxes on a rack. Aliquots also have a sample type.
Main constraints are:
Storing as few sample types per freezer as possible (i.e., ideal = all aliquots in Freezer1 are the same sample type)
Each sample draw event produces multiple aliquots. These aliquots from the same draw event should be split between two, and only two, freezers to prevent loss due to freezer failure.
Respect capacities of Racks and Boxes.
Each aliquot and box must be stored.
Storing a taller box in a rack requires removing a row divider in a rack. This results in that row only be used for taller boxes. Rows on a rack have 5 columns so for every 5 taller boxes, we loose capcity for 5 shorter boxes.
Domain Model
#PlanningEntity
public class Rack {
#InverseRelationShadowVariable(sourceVariableName = "box")
private List<Box> boxes = new ArrayList<>();
private Long boxCapacity;
}
#PlanningEntity
public class Box {
#PlanningVariable(valueRangeProviderRefs = "rackRange", nullable = true)
private Rack rack;
#InverseRelationShadowVariable(sourceVariableName = "aliquot")
private List<Aliquot> aliquots = new ArrayList<>();
private Long boxHeight;
private Long aliquotCapacity;
}
#PlanningEntity
public class Aliquot {
#PlanningVariable(valueRangeProviderRefs = "boxRange", nullable = true)
private Box box;
private Long aliquotHeight;
private SampleType sampleType;
}
public class SampleType {
private String name;
}
#PlanningSolution
public class AliquotMap {
#ValueRangeProvider(id = "rackRange")
#ProblemFactCollectionProperty
private List<Rack> rackList;
#ValueRangeProvider(id = "boxRange")
#PlanningEntityCollectionProperty
private List<Box> boxList;
#ValueRangeProvider(id = "aliquotRange")
#PlanningEntityCollectionProperty
private List<Aliquot> aliquotList;
}
This is the model I am proposing. My use of shadow variables and planning entities seems to make the model more complex than it needs to be though. Could this model be made without them or is there an entirely different modeling approach I should take?