diff --git a/app/parsers/concerns/bulkrax/csv_parser/csv_validation.rb b/app/parsers/concerns/bulkrax/csv_parser/csv_validation.rb
index f705c35a..cf040de6 100644
--- a/app/parsers/concerns/bulkrax/csv_parser/csv_validation.rb
+++ b/app/parsers/concerns/bulkrax/csv_parser/csv_validation.rb
@@ -76,9 +76,16 @@ def validate_csv(csv_file:, zip_file: nil, admin_set_id: nil) # rubocop:disable
           file_validator = CsvTemplate::FileValidator.new(csv_data, zip_file, admin_set_id)
 
           # 10. Item hierarchy for UI display
-          collections, works, file_sets = extract_validation_items(csv_data)
+          collections, works, file_sets = extract_validation_items(csv_data, all_ids)
 
           # 11. Assemble result
+          source_id_missing = !headers.map(&:to_s).include?(source_id_key.to_s)
+          if source_id_missing && Bulkrax.fill_in_blank_source_identifiers.blank?
+            all_models.each do |model|
+              missing_required << { model: model, field: source_id_key.to_s }
+            end
+          end
+
           row_errors  = validator_context[:errors]
           has_errors  = missing_required.any? || headers.blank? || csv_data.empty? ||
                         file_validator.missing_files.any? || row_errors.any?
@@ -189,14 +196,14 @@ def resolve_parent_split_pattern(mappings)
           split_val
         end
 
-        def extract_validation_items(csv_data) # rubocop:disable Metrics/MethodLength
+        def extract_validation_items(csv_data, all_ids = Set.new) # rubocop:disable Metrics/MethodLength
           child_to_parents = build_child_to_parents_map(csv_data)
           collections = []
           works       = []
           file_sets   = []
 
           csv_data.each do |item|
-            categorise_validation_item(item, child_to_parents, collections, works, file_sets)
+            categorise_validation_item(item, child_to_parents, all_ids, collections, works, file_sets)
           end
 
           [collections, works, file_sets]
@@ -205,6 +212,8 @@ def extract_validation_items(csv_data) # rubocop:disable Metrics/MethodLength
         def build_child_to_parents_map(csv_data)
           Hash.new { |h, k| h[k] = [] }.tap do |map|
             csv_data.each do |item|
+              next if item[:source_identifier].blank?
+
               parse_relationship_field(item[:children]).each do |child_id|
                 map[child_id] << item[:source_identifier]
               end
@@ -212,25 +221,25 @@ def build_child_to_parents_map(csv_data)
           end
         end
 
-        def categorise_validation_item(item, child_to_parents, collections, works, file_sets)
+        def categorise_validation_item(item, child_to_parents, all_ids, collections, works, file_sets) # rubocop:disable Metrics/ParameterLists
           item_id   = item[:source_identifier]
           title     = item[:raw_row]['title'] || item_id
           model_str = item[:model].to_s
 
           if model_str.casecmp('collection').zero? || model_str.casecmp('collectionresource').zero?
-            explicit = parse_relationship_field(item[:parent])
-            inferred = child_to_parents[item_id] || []
+            explicit = resolvable_ids(parse_relationship_field(item[:parent]), all_ids)
+            inferred = resolvable_ids(child_to_parents[item_id] || [], all_ids)
             collections << { id: item_id, title: title, type: 'collection',
                              parentIds: (explicit + inferred).uniq,
-                             childIds: parse_relationship_field(item[:children]) }
+                             childIds: resolvable_ids(parse_relationship_field(item[:children]), all_ids) }
           elsif model_str.casecmp('fileset').zero? || model_str.casecmp('hyrax::fileset').zero?
             file_sets << { id: item_id, title: title, type: 'file_set' }
           else
-            explicit = parse_relationship_field(item[:parent])
-            inferred = child_to_parents[item_id] || []
+            explicit = resolvable_ids(parse_relationship_field(item[:parent]), all_ids)
+            inferred = resolvable_ids(child_to_parents[item_id] || [], all_ids)
             works << { id: item_id, title: title, type: 'work',
                        parentIds: (explicit + inferred).uniq,
-                       childIds: parse_relationship_field(item[:children]) }
+                       childIds: resolvable_ids(parse_relationship_field(item[:children]), all_ids) }
           end
         end
 
@@ -239,6 +248,10 @@ def parse_relationship_field(value)
           value.to_s.split('|').map(&:strip).reject(&:blank?)
         end
 
+        def resolvable_ids(ids, all_ids)
+          ids.select { |id| all_ids.include?(id) }
+        end
+
         def apply_rights_statement_validation_override!(result, missing_required)
           only_rights = missing_required.present? &&
                         missing_required.all? { |h| h[:field].to_s == 'rights_statement' }
diff --git a/app/services/bulkrax/csv_template/explanation_builder.rb b/app/services/bulkrax/csv_template/explanation_builder.rb
index 7ecf08dc..2d64616e 100644
--- a/app/services/bulkrax/csv_template/explanation_builder.rb
+++ b/app/services/bulkrax/csv_template/explanation_builder.rb
@@ -21,7 +21,7 @@ def build_explanations(header_row)
       def build_explanation(column)
         mapping_key = @service.mapping_manager.mapped_to_key(column)
 
-        column_description = @descriptor.find_description_for(column)
+        column_description = source_identifier_description(column) || @descriptor.find_description_for(column)
         controlled_vocab_info = controlled_vocab_text(mapping_key)
         split_info = split_text(mapping_key, controlled_vocab_info)
 
@@ -34,6 +34,12 @@ def build_explanation(column)
         components.join("\n")
       end
 
+      def source_identifier_description(column)
+        return unless @service.mapping_manager.mapped_to_key(column) == 'source_identifier'
+        return if Bulkrax.fill_in_blank_source_identifiers.blank?
+        "Will be auto-generated if left blank.\nProviding one allows round-tripping and deduplication across imports."
+      end
+
       def controlled_vocab_text(field_name)
         vocab_terms = @service.field_analyzer.controlled_vocab_terms
         return unless vocab_terms.include?(field_name) || field_name == 'based_near'
diff --git a/app/validators/bulkrax/csv_row/duplicate_identifier.rb b/app/validators/bulkrax/csv_row/duplicate_identifier.rb
index d555731f..8aea9c85 100644
--- a/app/validators/bulkrax/csv_row/duplicate_identifier.rb
+++ b/app/validators/bulkrax/csv_row/duplicate_identifier.rb
@@ -8,6 +8,8 @@ module CsvRow
     module DuplicateIdentifier
       def self.call(record, row_index, context)
         source_id = record[:source_identifier]
+        return if source_id.blank? && Bulkrax.fill_in_blank_source_identifiers.present?
+
         source_id_label = context[:source_identifier] || 'source_identifier'
         first_row = context[:seen_ids][source_id]
 
diff --git a/spec/services/bulkrax/csv_template/csv_parser_template_spec.rb b/spec/services/bulkrax/csv_template/csv_parser_template_spec.rb
index bfdd8c88..19f05e69 100644
--- a/spec/services/bulkrax/csv_template/csv_parser_template_spec.rb
+++ b/spec/services/bulkrax/csv_template/csv_parser_template_spec.rb
@@ -158,6 +158,58 @@
         expect(result[:unrecognized].keys).to include('source_idenifier', 'titel', 'creater')
       end
     end
+
+    context 'source_identifier generation' do
+      context 'when fill_in_blank_source_identifiers is not configured' do
+        before { allow(Bulkrax).to receive(:fill_in_blank_source_identifiers).and_return(nil) }
+
+        it 'does not include source_identifier in missingRequired when the column is present' do
+          result = described_class.validate_csv(csv_file: csv_file, zip_file: nil)
+          source_id_missing = result[:missingRequired].any? { |h| h[:field].to_s == 'source_identifier' }
+          expect(source_id_missing).to be false
+        end
+
+        it 'treats a missing source_identifier column as an error' do
+          csv_without_source_id = Tempfile.new(['no_source_id', '.csv'])
+          csv_without_source_id.write("title,model\nTest Work,GenericWork\n")
+          csv_without_source_id.rewind
+
+          result = described_class.validate_csv(csv_file: csv_without_source_id, zip_file: nil)
+          expect(result[:isValid]).to be false
+          expect(result[:missingRequired]).to include(a_hash_including(field: 'source_identifier'))
+        ensure
+          csv_without_source_id.close
+          csv_without_source_id.unlink
+        end
+      end
+
+      context 'when fill_in_blank_source_identifiers is configured' do
+        before do
+          allow(Bulkrax).to receive(:fill_in_blank_source_identifiers)
+            .and_return(->(_parser, _index) { SecureRandom.uuid })
+        end
+
+        it 'does not include source_identifier in missingRequired even when the column is present' do
+          result = described_class.validate_csv(csv_file: csv_file, zip_file: nil)
+          source_id_missing = result[:missingRequired].any? { |h| h[:field].to_s == 'source_identifier' }
+          expect(source_id_missing).to be false
+        end
+
+        it 'does not treat a missing source_identifier column as an error' do
+          csv_without_source_id = Tempfile.new(['no_source_id', '.csv'])
+          csv_without_source_id.write("title,model\nTest Work,GenericWork\n")
+          csv_without_source_id.rewind
+
+          result = described_class.validate_csv(csv_file: csv_without_source_id, zip_file: nil)
+          expect(result[:isValid]).to be true
+          source_id_missing = result[:missingRequired].any? { |h| h[:field].to_s == 'source_identifier' }
+          expect(source_id_missing).to be false
+        ensure
+          csv_without_source_id.close
+          csv_without_source_id.unlink
+        end
+      end
+    end
   end
 
   describe 'TemplateContext' do
diff --git a/spec/validators/bulkrax/csv_row/duplicate_identifier_spec.rb b/spec/validators/bulkrax/csv_row/duplicate_identifier_spec.rb
index 11dd44cb..f346f6a3 100644
--- a/spec/validators/bulkrax/csv_row/duplicate_identifier_spec.rb
+++ b/spec/validators/bulkrax/csv_row/duplicate_identifier_spec.rb
@@ -29,4 +29,27 @@ def make_record(source_id)
     expect(error[:row]).to eq(3)
     expect(error[:value]).to eq('work1')
   end
+
+  context 'when fill_in_blank_source_identifiers is configured' do
+    before do
+      allow(Bulkrax).to receive(:fill_in_blank_source_identifiers)
+        .and_return(->(_parser, _index) { SecureRandom.uuid })
+    end
+
+    it 'skips the duplicate check for blank source_identifiers' do
+      context = make_context
+      described_class.call(make_record(nil), 2, context)
+      described_class.call(make_record(nil), 3, context)
+
+      expect(context[:errors]).to be_empty
+    end
+
+    it 'still records an error for duplicate non-blank identifiers' do
+      context = make_context
+      described_class.call(make_record('work1'), 2, context)
+      described_class.call(make_record('work1'), 3, context)
+
+      expect(context[:errors].length).to eq(1)
+    end
+  end
 end