diff --git a/.travis.yml b/.travis.yml
index 132be5d..f10b731 100644
--- a/.travis.yml
+++ b/.travis.yml
@@ -36,6 +36,7 @@ matrix:
 # command to install dependencies
 install:
     - python --version
+    - export PUBS_TESTS_MODE=ONLINE
 
 # command to run tests
 script: python setup.py test
diff --git a/tests/mock_requests.py b/tests/mock_requests.py
index 31e4ec3..a9ff9ab 100644
--- a/tests/mock_requests.py
+++ b/tests/mock_requests.py
@@ -1,3 +1,26 @@
+"""
+Mock the `requests.get` function, and handle collecting data to do so.
+
+Three modes are available, and controlled via the `PUBS_TESTS_MODE` environment
+variable. To modify the variable, under linux or macos, do one of:
+$ export PUBS_TESTS_MODE=MOCK
+$ export PUBS_TESTS_MODE=COLLECT
+$ export PUBS_TESTS_MODE=ONLINE
+
+The MOCK mode is the default one, active even if PUBS_TESTS_MODE has not been
+set. It use prefetched data to run pubs units tests relying on the `requests.get`
+function without the need of an internet connection (it is also much faster).
+The prefected data is save in the `test_apis_data.pickle` file.
+
+The COLLECT mode does real GET requests, and update the `test_apis_data.pickle`
+file. It is needed if you add or modify the test relying on `requests.get`.
+
+The ONLINE mode bypasses all this and use the original `requests.get` without
+accessing or updating the `test_apis_data.pickle` data. It might be useful when
+running tests on Travis for instance.
+"""
+
+
 import os
 try:
     import cPickle as pickle
@@ -6,9 +29,10 @@ except ImportError:
 
 import requests
 
-_orgininal_requests_get = requests.get
 
+_orgininal_requests_get = requests.get
 _collected_responses = []
+_data_filepath = os.path.join(os.path.dirname(__file__), 'test_apis_data.pickle')
 
 class MockingResponse:
     def __init__(self, text, status_code=200, error_msg=None):
@@ -26,14 +50,16 @@ mode = os.environ.get('PUBS_TESTS_MODE', 'MOCK')
 
 if mode == 'MOCK':
 
-    with open('test_apis_data.pickle', 'rb') as fd:
+    with open(os.path.join(_data_filepath), 'rb') as fd:
         _collected_responses.extend(pickle.load(fd))
 
     def mock_requests_get(*args, **kwargs):
         for args2, kwargs2, text, status_code, error_msg in _collected_responses:
             if args == args2 and kwargs == kwargs2:
                 return MockingResponse(text, status_code, error_msg)
-        raise KeyError('No stub data found for requests.get({}, {})'.format(args, kwargs))
+        raise KeyError(('No stub data found for requests.get({}, {}).\n You may'
+                        ' need to update the mock data. Look at the '
+                        'tests/mock_requests.py file for explanation').format(args, kwargs))
 
 elif mode == 'COLLECT':
 
@@ -48,13 +74,15 @@ elif mode == 'COLLECT':
         key = (sorted(args), sorted((k, v) for k, v in kwargs.items()))
 
         _collected_responses.append((args, kwargs, text, status_code, error_msg))
-        _save_collected_responses()
+        _save_collected_responses() # yes, we save everytime, because it's not
+                                    # clear how to run once after all the tests
+                                    # have run. If you figure it out...
 
         return MockingResponse(text, status_code, error_msg)
 
     def _save_collected_responses():
-        with open('test_apis_data.pickle', 'wb') as fd:
-            pickle.dump(_collected_responses, fd, protocol=3)
+        with open(os.path.join(_data_filepath), 'wb') as fd:
+            pickle.dump(_collected_responses, fd, protocol=2)
 
 elif mode == 'ONLINE':
     def mock_requests_get(*args, **kwargs):
diff --git a/test_apis_data.pickle b/tests/test_apis_data.pickle
similarity index 99%
rename from test_apis_data.pickle
rename to tests/test_apis_data.pickle
index a18f46a..1d7a914 100644
Binary files a/test_apis_data.pickle and b/tests/test_apis_data.pickle differ