Edit

In the RegistrationWorkingSetService.loadWorkingSetByReferenceID(Integer referenceID) a RegistrationWorkingSet is loaded with the following code:

List<String> REGISTRATION_INIT_STRATEGY = Arrays.asList(new String []{
            "blockedBy",
            // typeDesignation
            "typeDesignations.typeStatus",
            "typeDesignations.typifiedNames.typeDesignations", // important !!
            "typeDesignations.typeSpecimen",
            "typeDesignations.typeName.$",
            "typeDesignations.citation",
            "typeDesignations.citation.authorship.$",
            // name
            "name.$",
            "name.nomenclaturalReference.authorship.$",
            "name.nomenclaturalReference.inReference",
            "name.rank",
            "name.homotypicalGroup.typifiedNames",
            "name.status.type",
            "name.typeDesignations", // important !!"
            // institution
            "institution",
            }
    );

Reference reference = repo.getReferenceService().find(referenceID);
Pager<Registration> pager = repo.getRegistrationService().page(Optional.of(reference), null, null, null, REGISTRATION_INIT_STRATEGY);
return new RegistrationWorkingSet(makeDTOs(pager.getRecords()));

This usually works, but with a specific data set the pager contains Registrations where authorship
properties of the nomenclaturalReferences are not initializes even if the AdvancedBeanInitializer apparently is doing the initialization properly (also confirmed by debugging):

[eu.et.cd.pe.da.in.AdvancedBeanInitializer] -  processing /name.nomenclaturalReference
[eu.et.cd.pe.da.in.AdvancedBeanInitializer] -  invoke initialization on /name.nomenclaturalReference beans of class TaxonName ... 
[eu.et.cd.pe.da.in.AdvancedBeanInitializer] - bulk load /name.nomenclaturalReference
[eu.et.cd.pe.da.in.AdvancedBeanInitializer] - bulk load /name.nomenclaturalReference - DONE 
[eu.et.cd.pe.da.in.AdvancedBeanInitializer] -  processing /name.nomenclaturalReference.authorship
[eu.et.cd.pe.da.in.AdvancedBeanInitializer] -  invoke initialization on /name.nomenclaturalReference.authorship beans of class Reference ... 
[eu.et.cd.pe.da.in.AdvancedBeanInitializer] - bulk load /name.nomenclaturalReference.authorship
[eu.et.cd.pe.da.in.AdvancedBeanInitializer] - bulk load beans of class TeamOrPersonBase
[eu.et.cd.pe.da.in.AdvancedBeanInitializer] -  SELECT c FROM TeamOrPersonBase as c  WHERE c.id IN (:idSet) 
[eu.et.cd.pe.da.in.AdvancedBeanInitializer] - initialize bulk loaded beans of class TeamOrPersonBase: [1383]
[eu.et.cd.pe.da.in.AdvancedBeanInitializer] - bulk load - DONE

The actual initialization takes place in AdvancedBeanInitializer.bulkLoadLazyBeans(BeanInitNode node). Here a strange behavior can be observed in these special cases:

private void bulkLoadLazyBeans(node node) {
        for (Class<?> clazz : node.getLazyBeans().keySet()){
            Set<Serializable> idSet = node.getLazyBeans().get(clazz);
    String hql = " SELECT c FROM %s as c %s WHERE c.id IN (:idSet) ";
    hql = String.format(hql, clazz.getSimpleName(), autoInit.leftJoinFetch); // hql = "SELECT c FROM TeamOrPersonBase as c  WHERE c.id IN (:idSet)" 
    Query query = genericDao.getHqlQuery(hql);
    query.setParameterList("idSet", idSet);
    List<Object> list = query.list();

Now the TeamOrPersonBase entities are initialized.
However, checking the parent object contained in the node object passed as parameer reveals that the 'authorship' proxy in the object graph has not been initialized.

Preloading the reference in the RegistrationWorkingSetService.loadWorkingSetByReferenceID(Integer referenceID) method avoids the problem (in the below code only relevant lines are shown)

Reference reference = repo.getReferenceService().find(referenceID);
repo.getReferenceService().load(reference.getUuid());

Since this behavior is data dependent, I serialized the fully initialized registrations as loaded in the RegistrationDaoHibernateImpl.list(Optional<Reference> reference, Collection<RegistrationStatus> includedStatus, Integer limit, Integer start, List<String> propertyPaths). registrations-fully-initialized.ser can be used to restore the original data by de-serializing and persisting the objects to a cdm-database (version 4.15.0-SNAPSHOT) :

 try {
            FileInputStream fin = new FileInputStream("registrations-fully-initialized.ser");
            ObjectInputStream oin = new ObjectInputStream(fin);
            List<Registration> registrationsDeserialized = (List<Registration>) oin.readObject();
        } catch (ClassNotFoundException | IOException e) {
            // TODO Auto-generated catch block
            e.printStackTrace();
        }

NOTE: Doing the serialization in RegistrationDaoHibernateImpl.list() also removed the problem described in here:

        try {
            FileOutputStream fout = new FileOutputStream("registrations-before-initialization.ser");
            ObjectOutputStream oos = new ObjectOutputStream(fout);
            oos.writeObject(results);
        } catch (IOException e) {
            // TODO Auto-generated catch block
            e.printStackTrace();
        }

        defaultBeanInitializer.initializeAll(results, propertyPaths);

        // initialization is incomplete

        try {
            FileOutputStream fout = new FileOutputStream("registrations-after-initialization.ser");
            ObjectOutputStream oos = new ObjectOutputStream(fout);
            oos.writeObject(results);
        } catch (IOException e) {
            // TODO Auto-generated catch block
            e.printStackTrace();
        }

        // fully initialized now! the file registrations-fully-initialized.ser has been created after this