diff --git a/TODO b/TODO
index b306082..bb1357c 100644
--- a/TODO
+++ b/TODO
@@ -1,6 +1,5 @@
 - add tab-completion for label during motivation grading
 - add common labels to criteria.txt
 - tab-completion does not work out of the box in the Mac terminal: https://github.com/ASPP/grader/issues/26
-- if two reviewers add the same label there is a git-conflict
 - add a `o` key to the grade verb that allows to set overrides. Fields that can be overridden must be tab-auto-completed and possible values for the field must be also tab-auto-completed.
 - verify that we can dynamically change the formula and this has an effect on the ranking
diff --git a/grader/applications.py b/grader/applications.py
index 95ae1ec..f499cb2 100644
--- a/grader/applications.py
+++ b/grader/applications.py
@@ -1,11 +1,13 @@
 import collections
 import configparser
+import contextlib
 import csv
 from fnmatch import fnmatch
 import functools
 import io
 import itertools
 import keyword
+import locale
 import math
 import pprint
 import re
@@ -15,6 +17,17 @@
 from . import (person, vector, util)
 from .util import printff
 
+
+@contextlib.contextmanager
+def override_locale(category, locale_string):
+    # Context manager to set locale temporarily.
+    # Strangely, it seems that there is no builtin that does that.
+    prev_locale_string = locale.getlocale(category)
+    locale.setlocale(category, locale_string)
+    yield
+    locale.setlocale(category, prev_locale_string)
+
+
 DEBUG_MAPPINGS = False
 
 # CSV-file:
@@ -376,18 +389,41 @@ def get_ratings(self, field):
 
     def save(self, filename=None):
         filename = filename or self.filename
-        with open(filename, 'wt') as file:
-            self.save_to_file(file)
+        text = self.to_string()
 
-    def save_to_file(self, file):
+        filename.write_text(text)
+
+        printff(f'Saved changes to {filename}')
+
+    def to_string(self):
         # save our data to the INI file
         cp = configparser.ConfigParser(comment_prefixes='#', inline_comment_prefixes='#')
-        cp.read_dict(self.data)
-        cp.write(file)
-
-        name = getattr(file, 'name', '(tmp)')
-        printff(f'Saved changes to {name}')
 
+        with override_locale(locale.LC_COLLATE, 'en_US.UTF-8'):
+            # The order of sections is kept stable.
+            #
+            # Within a section, we sort the fields alphabetically,
+            # with the en_US.UTF-8 collation, which puts accented
+            # letters in the expected place, usually right after the
+            # unaccented version.
+            #
+            # As an exception, the [*_rating] sections are not sorted,
+            # so that the items remain from "lowest" to "highest".
+            sorted_data = {
+                name:{
+                    k:values[k] for k in sorted(
+                        values,
+                        key=(lambda x:0) if name.endswith('_rating') else functools.cmp_to_key(locale.strcoll),
+                    )
+                }
+                for name, values in self.data.items()
+            }
+
+        out = io.StringIO()
+
+        cp.read_dict(sorted_data)
+        cp.write(out)
+        return out.getvalue()
 
     def __setitem__(self, key, value):
         # allow to set items in the section of the INI using a dotted form, for ex:
diff --git a/grader/test_applications.py b/grader/test_applications.py
index 7831112..f1aaa05 100644
--- a/grader/test_applications.py
+++ b/grader/test_applications.py
@@ -1,4 +1,3 @@
-import io
 import pathlib
 import os
 import time
@@ -50,6 +49,11 @@ def test_all_years(path):
 person two = 1
 some son jr. = 1
 
+[verify_rating_is_unsorted_rating]
+novice = +10.0
+competent = 0.0
+expert = -10.0
+
 [labels]
 john doe = VEGAN, VIP
 jędrzej marcin mirosławski piołun = UTF-8, VEGAN
@@ -57,6 +61,36 @@ def test_all_years(path):
 
 """
 
+ini_sorted = """\
+[extra]
+key_num = 111.5
+key_str = value
+
+[motivation_score-zbyszek]
+jędrzej marcin mirosławski piołun = 1
+person one = 1
+person three = 0
+person two = -1
+some son jr. = -1
+
+[motivation_score-other]
+person one = -1
+person two = 1
+some son jr. = 1
+
+[verify_rating_is_unsorted_rating]
+novice = 10.0
+competent = 0.0
+expert = -10.0
+
+[labels]
+jędrzej marcin mirosławski piołun = UTF-8, VEGAN
+john doe = VEGAN, VIP
+person one = PALEO
+
+"""
+
+
 def get_ini(tmp_path, *extra, ini_filename='ini1.ini'):
     input = tmp_path / ini_filename
     input.write_text(ini_string + '\n'.join(extra))
@@ -167,7 +201,7 @@ def test_applications_ini_save(tmp_path):
     out = tmp_path / 'ini1.copy'
     ini.save(out)
 
-    assert out.read_text() == ini_string
+    assert out.read_text() == ini_sorted
 
     # Replace an exisiting entry.
     # We use an int, but the type is converted to float internally
@@ -292,11 +326,10 @@ def test_applications_labels(app):
     assert app['Person One'].remove_label('PALEO') is True
     assert app['Person One'].labels == []
 
-    out = io.StringIO()
-    app.ini.save_to_file(file=out)
+    text = app.ini.to_string()
 
-    assert 'john doe = VEGAN, VIP' in out.getvalue()
-    assert 'jędrzej marcin mirosławski piołun = UTF-8, VEGAN' in out.getvalue()
+    assert 'john doe = VEGAN, VIP' in text
+    assert 'jędrzej marcin mirosławski piołun = UTF-8, VEGAN' in text
 
 def test_applications_all_labels(app):
     assert app.all_labels() == {'PALEO', 'UTF-8', 'VEGAN'}