diff --git a/db/migrate/20160720185407_clean_inconsistencies.rb b/db/migrate/20160720185407_clean_inconsistencies.rb
index 8732cf8..d99fd20 100644
--- a/db/migrate/20160720185407_clean_inconsistencies.rb
+++ b/db/migrate/20160720185407_clean_inconsistencies.rb
@@ -1,5 +1,34 @@
+require 'fileutils'
+
 class CleanInconsistencies < ActiveRecord::Migration
-  def self.up
+  def backup_dir
+    return @backup_dir if @backup_dir
+    @backup_dir = Rails.root.join('db', 'backup', Time.now.strftime('%Y-%m-%d_%H-%M-%S'))
+    FileUtils.mkdir_p(@backup_dir)
+  end
+
+  def backup(name, query, header: true)
+    say_with_time("backup(#{name.inspect}, #{query.inspect})") do
+      copy_query = "COPY (#{query}) TO STDOUT WITH DELIMITER ',' CSV #{header ? 'HEADER' : ''}"
+
+      File.open(backup_dir.join("#{name}.csv"), 'a') do |f|
+        connection.raw_connection.copy_data(copy_query) do
+          while line = connection.raw_connection.get_copy_data
+            f.write(line)
+          end
+        end
+      end
+    end
+  end
+
+  def backup_and_delete_missing(table, exists_query)
+    backup(table, "SELECT * FROM \"#{table}\" WHERE NOT EXISTS(#{exists_query})")
+    execute "DELETE FROM \"#{table}\" WHERE NOT EXISTS(#{exists_query})"
+  end
+
+  def up
+    say "WARNING: destructive migration necessary. Deleted data will be backed up to #{backup_dir}"
+
     # Unset project reference for repositories with non-existing projects
     execute <<-SQL
       UPDATE repositories AS r
@@ -10,61 +39,62 @@ def self.up
     SQL
 
     # Delete processings with non-existing repositories
-    execute <<-SQL
-      DELETE FROM processings AS p
-      WHERE NOT EXISTS(
-        SELECT 1 FROM repositories AS r WHERE r.id = p.repository_id
-      )
-    SQL
+    backup_and_delete_missing("processings",
+      "SELECT 1 FROM repositories AS r WHERE r.id = processings.repository_id")
 
     # Delete process times with non-existing processings
-    execute <<-SQL
-      DELETE FROM process_times AS t
-      WHERE NOT EXISTS (
-        SELECT 1 FROM processings AS p WHERE p.id = t.processing_id
-      )
-    SQL
+    backup_and_delete_missing("process_times",
+      "SELECT 1 FROM processings AS p WHERE p.id = process_times.processing_id")
 
     # Delete module results with non-existing processings
-    execute <<-SQL
-      DELETE FROM module_results AS m
-      WHERE NOT EXISTS (
-        SELECT 1 FROM processings AS p WHERE p.id = m.processing_id
-      )
-    SQL
+    backup_and_delete_missing("module_results",
+      "SELECT 1 FROM processings AS p WHERE p.id = module_results.processing_id")
 
     # Delete kalibro modules with non-existing module results
+    backup_and_delete_missing("kalibro_modules",
+      "SELECT 1 FROM module_results AS m WHERE m.id = kalibro_modules.module_result_id")
+
+    # Fix up metric results type, even before backing up so the backup is cleaner
     execute <<-SQL
-      DELETE FROM kalibro_modules AS k
-      WHERE NOT EXISTS (
-        SELECT 1 FROM module_results AS m WHERE m.id = k.module_result_id
-      )
+      UPDATE metric_results SET "type" = 'TreeMetricResult' WHERE "type" = 'MetricResult'
     SQL
 
     # Delete metric results with non-existing module results
-    execute <<-SQL
-      DELETE FROM metric_results AS met
-      WHERE NOT EXISTS (
-        SELECT 1 FROM module_results AS mod WHERE mod.id = met.module_result_id
-      )
-    SQL
+    backup_and_delete_missing("metric_results",
+      "SELECT 1 FROM module_results AS m WHERE m.id = metric_results.module_result_id")
 
-    # Delete duplicate metric_results. Group them by (module_result, metric_configuration),
-    # then delete all but the one with the highest ID
-    # The double wrapping on the inner query is necessary because window functions
-    # cannot be used in WHERE in PostgreSQL.
-    execute <<-SQL
-      DELETE FROM metric_results
-      WHERE id IN (
-        SELECT t.id FROM (
-          SELECT id, ROW_NUMBER() OVER (PARTITION BY module_result_id, metric_configuration_id, "type"
-                                        ORDER BY id DESC) AS rnum
-          FROM metric_results
-          WHERE "type" = 'TreeMetricResult'
-        ) AS t
-        WHERE t.rnum > 1
-      )
+    # Delete duplicate metric_results. Group them by (module_result_id, metric_configuration_id),
+    # then delete all but the one with the highest ID. The double wrapping on the inner query is
+    # necessary because window functions cannot be used in WHERE in PostgreSQL.
+    repeated_metric_result_query = exec_query <<-SQL
+      SELECT t.id FROM (
+        SELECT metric_results.*, ROW_NUMBER() OVER (
+          PARTITION BY module_result_id, metric_configuration_id, "type"
+          ORDER BY id DESC) AS rnum
+        FROM metric_results
+        WHERE "type" = 'TreeMetricResult'
+      ) AS t
+      WHERE t.rnum > 1
     SQL
+
+    unless repeated_metric_result_query.empty?
+      repeated_metric_result_ids = repeated_metric_result_query.rows.flat_map(&:first).join(',')
+
+      # Replace default messages with custom ones to avoid flooding the screen with the huge query
+      say_with_time('backup("metric_results", "SELECT * metric_results WHERE id IN (...)")') do
+        suppress_messages do
+          backup('metric_results',
+            "SELECT * FROM metric_results WHERE id IN (#{repeated_metric_result_ids})",
+            header: false)
+        end
+      end
+
+      say_with_time('execute("DELETE FROM metric_results WHERE id IN (...)")') do
+        suppress_messages do
+          execute "DELETE FROM metric_results WHERE id IN (#{repeated_metric_result_ids})"
+        end
+      end
+    end
   end
 
   def self.down