diff --git a/runbot_merge/models/stagings_create.py b/runbot_merge/models/stagings_create.py
index 231ee969..017b63a9 100644
--- a/runbot_merge/models/stagings_create.py
+++ b/runbot_merge/models/stagings_create.py
@@ -61,8 +61,8 @@ def try_staging(branch: Branch) -> Optional[Stagings]:
     def log(label: str, batches: Batch) -> None:
         _logger.info(label, ', '.join(batches.mapped('prs.display_name')))
 
-    alone, batches = ready_prs(for_branch=branch)
-    print('alone', alone, 'ready', batches)
+    alone, batches = ready_batches(for_branch=branch)
+
     if alone:
         log("staging high-priority PRs %s", batches)
     elif branch.project_id.staging_priority == 'default':
@@ -75,13 +75,6 @@ def try_staging(branch: Branch) -> Optional[Stagings]:
             # first as long as there's room
             log("staging ready PRs %s (prioritising splits)", batches)
     elif branch.project_id.staging_priority == 'ready':
-        # splits are ready by definition, we need to exclude them from the
-        # ready rows otherwise we'll re-stage the splits so if an error is legit
-        # we cycle forever
-        # FIXME: once the batches are less shit, store this durably on the
-        #        batches and filter out when fetching readies (?)
-        batches -= branch.split_ids.batch_ids
-
         if batches:
             log("staging ready PRs %s (prioritising ready)", batches)
         else:
@@ -91,11 +84,6 @@ def try_staging(branch: Branch) -> Optional[Stagings]:
             log("staging split PRs %s (prioritising ready)", batches)
     else:
         assert branch.project_id.staging_priority == 'largest'
-        # splits are ready by definition, we need to exclude them from the
-        # ready rows otherwise ready always wins but we re-stage the splits, so
-        # if an error is legit we'll cycle forever
-        batches -= branch.split_ids.batch_ids
-
         maxsplit = max(branch.split_ids, key=lambda s: len(s.batch_ids), default=branch.env['runbot_merge.split'])
         _logger.info("largest split = %d, ready = %d", len(maxsplit.batch_ids), len(batches))
         # bias towards splits if len(ready) = len(batch_ids)
@@ -201,13 +189,17 @@ For-Commit-Id: {it.head}
     return st
 
 
-def ready_prs(for_branch: Branch) -> Tuple[bool, Batch]:
+def ready_batches(for_branch: Branch) -> Tuple[bool, Batch]:
     env = for_branch.env
+    # splits are ready by definition, we need to exclude them from the ready
+    # rows otherwise if a prioritised (alone) PR is part of a split it'll be
+    # staged through priority *and* through split.
+    split_ids = for_branch.split_ids.batch_ids.ids
     env.cr.execute("""
         SELECT max(priority)
         FROM runbot_merge_batch
-        WHERE blocked IS NULL AND target = %s
-    """, [for_branch.id])
+        WHERE blocked IS NULL AND target = %s AND NOT id = any(%s)
+    """, [for_branch.id, split_ids])
     alone = env.cr.fetchone()[0] == 'alone'
 
     return (
@@ -216,6 +208,7 @@ def ready_prs(for_branch: Branch) -> Tuple[bool, Batch]:
             ('target', '=', for_branch.id),
             ('blocked', '=', False),
             ('priority', '=', 'alone') if alone else (1, '=', 1),
+            ('id', 'not in', split_ids),
         ], order="priority DESC, id ASC"),
     )
 
diff --git a/runbot_merge/tests/test_basic.py b/runbot_merge/tests/test_basic.py
index 273c0aae..dc458b34 100644
--- a/runbot_merge/tests/test_basic.py
+++ b/runbot_merge/tests/test_basic.py
@@ -2969,6 +2969,50 @@ class TestBatching(object):
         assert not p_01.staging_id, "p_01 should not be picked up as it's failed"
         assert p_21.staging_id, "p_21 should have been staged"
 
+    def test_urgent_split(self, env, repo, config):
+        """Ensure that urgent (alone) PRs which get split don't get
+        double-merged
+        """
+        with repo:
+            repo.make_commits(
+                None,
+                Commit("initial", tree={'a': '1'}),
+                ref="heads/master"
+            )
+
+            pr01 = self._pr(
+                repo, "PR1", [{'b': '1'}],
+                user=config['role_user']['token'],
+                reviewer=None,
+            )
+            pr01.post_comment('hansen alone r+', config['role_reviewer']['token'])
+            pr02 = self._pr(
+                repo, "PR2", [{'c': '1'}],
+                user=config['role_user']['token'],
+                reviewer=None,
+            )
+            pr02.post_comment('hansen alone r+', config['role_reviewer']['token'])
+        env.run_crons('runbot_merge.process_updated_commits')
+        pr01_id = to_pr(env, pr01)
+        assert pr01_id.blocked is False
+        pr02_id = to_pr(env, pr02)
+        assert pr01_id.blocked is False
+
+        env.run_crons()
+        st = pr01_id.staging_id
+        assert st and pr02_id.staging_id == st
+        with repo:
+            repo.post_status('staging.master', 'failure', 'ci/runbot')
+        env.run_crons()
+        # should have cancelled the staging, split it, and re-staged the first
+        # half of the split
+        assert st.state == 'failure'
+        assert pr01_id.staging_id and pr01_id.staging_id != st
+        assert not pr02_id.staging_id
+        split_prs = env['runbot_merge.split'].search([]).batch_ids.prs
+        assert split_prs == pr02_id, \
+            f"only the unstaged PR {pr02_id} should be in a split, found {split_prs}"
+
     @pytest.mark.skip(reason="Maybe nothing to do, the PR is just skipped and put in error?")
     def test_batching_merge_failure(self):
         pass