diff --git a/swh/loader/pypi/loader.py b/swh/loader/pypi/loader.py
--- a/swh/loader/pypi/loader.py
+++ b/swh/loader/pypi/loader.py
@@ -106,13 +106,16 @@
             list of (filename, sha256) tuples.
 
         """
+        if not last_snapshot or 'branches' not in last_snapshot:
+            return {}
+
         revs = [rev['target'] for rev in last_snapshot['branches'].values()]
         known_revisions = self.storage.revision_get(revs)
-        ret = []
+        ret = {}
         for revision in known_revisions:
             if 'original_artifact' in revision['metadata']:
                 artifact = revision['metadata']['original_artifact']
-                ret.append((artifact['filename'], artifact['sha256']))
+                ret[artifact['filename'], artifact['sha256']] = revision['id']
         return ret
 
     def _last_snapshot(self):
@@ -146,18 +149,11 @@
 
         """
         last_snapshot = self._last_snapshot()
-        if last_snapshot:
-            self._snapshot = last_snapshot.copy()
-            known_artifacts = self._known_artifacts(self._snapshot)
-        else:
-            self._snapshot = {
-                'branches': {}
-            }
-            known_artifacts = []
+        self.known_artifacts = self._known_artifacts(last_snapshot)
         # and the artifacts
         # that will be the source of data to retrieve
-        self.release_artifacts = self.project.download_new_releases(
-            known_artifacts
+        self.new_artifacts = self.project.download_new_releases(
+            self.known_artifacts
         )
         # temporary state
         self._contents = []
@@ -182,7 +178,7 @@
             return False
 
         try:
-            data = next(self.release_artifacts)
+            data = next(self.new_artifacts)
         except StopIteration:
             self.done = True
             return False
@@ -223,18 +219,34 @@
             revision_identifier(_revision))
         self._revisions.append(_revision)
 
-        branch_name = artifact['filename'].encode('utf-8')
-        self._snapshot['branches'][branch_name] = {
-            'target': _revision['id'],
-            'target_type': 'revision',
-        }
+        artifact_key = artifact['filename'], artifact['sha256']
+        self.known_artifacts[artifact_key] = _revision['id']
 
         return not self.done
 
+    def target_from_artifact(self, filename, sha256):
+        target = self.known_artifacts.get((filename, sha256))
+        if target:
+            return {
+                'target': target,
+                'target_type': 'revision',
+            }
+        return None
+
     def generate_and_load_snapshot(self):
-        self._snapshot['id'] = identifier_to_bytes(
-            snapshot_identifier(self._snapshot))
-        self.maybe_load_snapshot(self._snapshot)
+        branches = {}
+        for release, artifacts in self.project.all_release_artifacts().items():
+            for filename, sha256 in artifacts:
+                branch_name = filename
+                target = self.target_from_artifact(filename, sha256)
+                branches[branch_name.encode('utf-8')] = target
+
+        snapshot = {
+            'branches': branches,
+        }
+        snapshot['id'] = identifier_to_bytes(
+            snapshot_identifier(snapshot))
+        self.maybe_load_snapshot(snapshot)
 
     def store_data(self):
         """(override) This sends collected objects to storage.
diff --git a/swh/loader/pypi/tests/test_loader.py b/swh/loader/pypi/tests/test_loader.py
--- a/swh/loader/pypi/tests/test_loader.py
+++ b/swh/loader/pypi/tests/test_loader.py
@@ -217,16 +217,16 @@
 
     def _known_artifacts(self, last_snapshot):
         """List corresponding seen release artifacts"""
-        yield from [
+        return {
             (
                 '0805nexter-1.1.0.zip',
                 '52cd128ad3afe539478abc7440d4b043384295fbe6b0958a237cb6d926465035'  # noqa
-            ),
+            ): b'L\x99\x89\x1f\x93\xb8\x14P8Ww#Z7\xb5\xe9f\xdd\x15q',
             (
                 '0805nexter-1.2.0.zip',
                 '49785c6ae39ea511b3c253d7621c0b1b6228be2f965aca8a491e6b84126d0709'  # noqa
-            )
-        ]
+            ): b'\xe4E\xdaM\xa2+1\xbf\xeb\xb6\xff\xc48=\xbf\x83\x9a\x07M!',
+        }
 
 
 class LoaderNoNewChangesSinceLastVisitITest(BaseLoaderITest):
@@ -374,21 +374,21 @@
         }
 
     def _known_artifacts(self, last_snapshot):
-        """List corresponding seen release artifacts"""
-        return [
+        """Map previously seen release artifacts to their revision"""
+        return {
             (
                 '0805nexter-1.1.0.zip',
                 '52cd128ad3afe539478abc7440d4b043384295fbe6b0958a237cb6d926465035'  # noqa
-            ),
+            ): b'L\x99\x89\x1f\x93\xb8\x14P8Ww#Z7\xb5\xe9f\xdd\x15q',
             (
                 '0805nexter-1.2.0.zip',
                 '49785c6ae39ea511b3c253d7621c0b1b6228be2f965aca8a491e6b84126d0709'  # noqa
-            ),
+            ): b'\xe4E\xdaM\xa2+1\xbf\xeb\xb6\xff\xc48=\xbf\x83\x9a\x07M!',
             (
                 '0805nexter-1.3.0.zip',
                 '7097c49fb8ec24a7aaab54c3dbfbb5a6ca1431419d9ee0f6c363d9ad01d2b8b1'  # noqa
-            )
-        ]
+            ): b'\xfbF\xe4\x96\x05\xb0\xbb\xe6\x9f\x8cS\xd3\x15\xe8\x93p\xe7\xc6\xcb]',  # noqa
+        }
 
 
 class LoaderChangesOldReleaseArtifactRemovedSinceLastVisit(BaseLoaderITest):
@@ -399,14 +399,6 @@
     - a new release has been uploaded
     - an older one has been removed
 
-    # What wrongly happens now
-
-    The visit results in a new snapshot, such snapshot shares the same
-    branches as last visit's snapshot (including the unpublished
-    release artifacts).
-
-    # What should happen:
-
     The visit should result in a new snapshot. Such snapshot shares some of
     the same branches as prior visit (but not all):
 
@@ -458,19 +450,21 @@
         self.assertDirectoriesOk(expected_directories)
 
         expected_revisions = {
+            # 1.4.0
             '5e91875f096ac48c98d74acf307439a3490f2827': '770e21215ecac53cea331d8ea4dc0ffc9d979367',  # noqa
         }
         self.assertRevisionsOk(expected_revisions)
 
         old_revisions = {
-            '4c99891f93b81450385777235a37b5e966dd1571': '05219ba38bc542d4345d5638af1ed56c7d43ca7d',  # noqa
+            # 1.2.0
             'e445da4da22b31bfebb6ffc4383dbf839a074d21': 'b178b66bd22383d5f16f4f5c923d39ca798861b4',  # noqa
+            # 1.3.0
             'fb46e49605b0bbe69f8c53d315e89370e7c6cb5d': 'e226e7e4ad03b4fc1403d69a18ebdd6f2edd2b3a',  # noqa
         }
         for rev, dir_id in old_revisions.items():
             expected_revisions[rev] = dir_id
 
-        expected_snapshot_id = '0f36392a2c825bafa766b371f2485f9549eec41c'
+        expected_snapshot_id = 'fb192f35397812776377fa758e0ba4cf20a4cf5d'
         self.assertSnapshotOk(expected_snapshot_id, expected_revisions)
 
         _id = hashutil.hash_to_hex(self.loader._last_snapshot()['id'])