diff --git a/olasd/consistency/.gitignore b/olasd/consistency/.gitignore
new file mode 100644
index 0000000..f3770f8
--- /dev/null
+++ b/olasd/consistency/.gitignore
@@ -0,0 +1 @@
+broken_*
diff --git a/olasd/consistency/check_swh.py b/olasd/consistency/check_swh.py
new file mode 100755
index 0000000..ee4b847
--- /dev/null
+++ b/olasd/consistency/check_swh.py
@@ -0,0 +1,122 @@
+#!/usr/bin/python3
+
+import os
+import threading
+
+import psycopg2
+
+import converters
+from swh.storage import converters as db_converters
+from swh.model import identifiers
+
+id_to_str = identifiers.identifier_to_str
+
+DSN = ('host=somerset.internal.softwareheritage.org port=5433 '
+       'user=guest dbname=softwareheritage')
+
+REVISION_COLUMNS = [
+    ("r.id", "id", converters.tobytes),
+    ("date", "date", converters.todate),
+    ("date_offset", "date_offset", converters.toint),
+    ("committer_date", "committer_date", converters.todate),
+    ("committer_date_offset", "committer_date_offset", converters.toint),
+    ("type", "type", converters.tostr),
+    ("directory", "directory", converters.tobytes),
+    ("message", "message", converters.tobytes),
+    ("synthetic", "synthetic", converters.tobool),
+    ("metadata", "metadata", converters.tojson),
+    ("date_neg_utc_offset", "date_neg_utc_offset", converters.tobool),
+    ("committer_date_neg_utc_offset", "committer_date_neg_utc_offset",
+     converters.tobool),
+    ("array(select parent_id::bytea from revision_history rh "
+     "where rh.id = r.id order by rh.parent_rank asc)",
+     "parents", converters.tolist),
+    ("a.id", "author_id", converters.toint),
+    ("a.name", "author_name", converters.tobytes),
+    ("a.email", "author_email", converters.tobytes),
+    ("a.fullname", "author_fullname", converters.tobytes),
+    ("c.id", "committer_id", converters.toint),
+    ("c.name", "committer_name", converters.tobytes),
+    ("c.email", "committer_email", converters.tobytes),
+    ("c.fullname", "committer_fullname", converters.tobytes),
+]
+
+REVISION_QUERY = '''
+copy (
+    select %s
+    from revision r
+    left join person a on r.author = a.id
+    left join person c on r.committer = c.id
+) to stdout
+''' % ', '.join('%s as %s' % (column, alias)
+                for column, alias, _ in REVISION_COLUMNS)
+
+RELEASE_COLUMNS = [
+    ("r.id", "id", converters.tobytes),
+    ("date", "date", converters.todate),
+    ("date_offset", "date_offset", converters.toint),
+    ("comment", "comment", converters.tobytes),
+    ("r.name", "name", converters.tobytes),
+    ("synthetic", "synthetic", converters.tobool),
+    ("date_neg_utc_offset", "date_neg_utc_offset", converters.tobool),
+    ("target", "target", converters.tobytes),
+    ("target_type", "target_type", converters.tostr),
+    ("a.id", "author_id", converters.toint),
+    ("a.name", "author_name", converters.tobytes),
+    ("a.email", "author_email", converters.tobytes),
+    ("a.fullname", "author_fullname", converters.tobytes),
+]
+
+RELEASE_QUERY = '''
+copy (
+    select %s
+    from release r
+    left join person a on r.author = a.id
+) to stdout
+''' % ', '.join('%s as %s' % (column, alias)
+                for column, alias, _ in RELEASE_COLUMNS)
+
+
+def process_query(query, columns, db_converter):
+    r_fd, w_fd = os.pipe()
+    db = psycopg2.connect(DSN)
+
+    def get_data_thread():
+        cursor = db.cursor()
+        cursor.copy_expert(query, open(w_fd, 'wb'))
+        cursor.close()
+        db.commit()
+
+    data_thread = threading.Thread(target=get_data_thread)
+    data_thread.start()
+
+    r = open(r_fd, 'rb')
+    for line in r:
+        fields = {
+            alias: decoder(value)
+            for (_, alias, decoder), value
+            in zip(columns, line[:-1].decode('utf-8').split('\t'))
+        }
+        yield db_converter(fields)
+
+    r.close()
+
+    data_thread.join()
+
+
+if __name__ == '__main__':
+    with open('broken_releases', 'w') as broken_releases:
+        for release in process_query(RELEASE_QUERY, RELEASE_COLUMNS,
+                                     db_converters.db_to_release):
+            intrinsic_id = id_to_str(release['id'])
+            computed_id = id_to_str(identifiers.release_identifier(release))
+            if intrinsic_id != computed_id:
+                print(intrinsic_id, computed_id, file=broken_releases)
+
+    with open('broken_revisions', 'w') as broken_revisions:
+        for revision in process_query(REVISION_QUERY, REVISION_COLUMNS,
+                                      db_converters.db_to_revision):
+            intrinsic_id = id_to_str(revision['id'])
+            computed_id = id_to_str(identifiers.revision_identifier(revision))
+            if intrinsic_id != computed_id:
+                print(intrinsic_id, computed_id, file=broken_revisions)
diff --git a/olasd/consistency/converters.py b/olasd/consistency/converters.py
new file mode 100644
index 0000000..c72b377
--- /dev/null
+++ b/olasd/consistency/converters.py
@@ -0,0 +1,59 @@
+import json
+import dateutil
+
+
+def unescape(value):
+    """PostgreSQL text format unescape"""
+    return value.replace('\\\\', '\\')
+
+
+def default_converter(type):
+    def inner_converter(value, type=type):
+        if value == r'\N':
+            return None
+        else:
+            return type(unescape(value))
+
+    return inner_converter
+
+
+def tobool(value):
+    return {
+        '\\N': None,
+        't': True,
+        'f': False
+    }[value]
+
+
+def tobytes(value):
+    if value == r'\N':
+        return None
+    value = unescape(value)
+    if not value.startswith(r'\x'):
+        raise ValueError('Not an escaped bytea: %r' % value)
+    return bytes.fromhex(value[2:])
+
+
+def tojson(value):
+    if value == r'\N':
+        return None
+    return json.loads(unescape(value))
+
+
+def todate(value):
+    if value == r'\N':
+        return None
+    return dateutil.parser.parse(unescape(value))
+
+
+def tolist(value):
+    if value == r'\N':
+        return None
+    value = value[1:-1]
+    return [tobytes(unescape(item[1:-1]))
+            for item in (value.split(',')
+                         if value else [])]
+
+
+toint = default_converter(int)
+tostr = default_converter(str)