diff --git a/src/test/recovery/t/long_fork_reproducer.pl b/src/test/recovery/t/long_fork_reproducer.pl
new file mode 100644
index 00000000000..319679a25f7
--- /dev/null
+++ b/src/test/recovery/t/long_fork_reproducer.pl
@@ -0,0 +1,112 @@
+# Reproduce the Long Fork anomaly between a primary and standby.
+# This is a standalone reproducer, not an entry in the recovery test schedule.
+#
+# Scenario:
+#   * t0: x=0, y=0 on primary, standby1, and standby2.
+#   * Pause replay on standby1.
+#   * tx1 (synchronous_commit=remote_apply): UPDATE x=1; COMMIT. Waits on pause.
+#   * Standby2 receives tx1's WAL and applies it.  We poll until standby2 has
+#     reached the primary's current flush LSN, then pause standby2.  Standby2
+#     now holds at (x=1, y=0).
+#   * tx2 (synchronous_commit=off): UPDATE y=1; COMMIT.  Returns immediately.
+#     tx2's WAL ships to both standbys but neither applies it (both paused).
+#   * Read primary: a fresh snapshot sees tx1 as running and tx2 as committed,
+#     returning (x=0, y=1).
+#   * Read standby2: paused at the post-tx1 LSN, returning (x=1, y=0).
+#
+use strict;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
+use Test::More;
+
+my $primary = PostgreSQL::Test::Cluster->new('primary');
+$primary->init(allows_streaming => 1);
+$primary->start;
+$primary->backup('bk');
+
+my $standby1 = PostgreSQL::Test::Cluster->new('standby1');
+$standby1->init_from_backup($primary, 'bk', has_streaming => 1);
+$standby1->start;
+
+my $standby2 = PostgreSQL::Test::Cluster->new('standby2');
+$standby2->init_from_backup($primary, 'bk', has_streaming => 1);
+$standby2->start;
+
+$primary->safe_psql('postgres',
+	"ALTER SYSTEM SET synchronous_standby_names = 'standby1';");
+$primary->reload;
+
+$primary->poll_query_until('postgres', q[
+    SELECT sync_state = 'sync'
+    FROM pg_stat_replication
+    WHERE application_name = 'standby1'
+]) or die "standby1 never reached sync_state='sync'";
+
+$primary->poll_query_until('postgres', q[
+    SELECT count(*) = 1
+    FROM pg_stat_replication
+    WHERE application_name = 'standby2'
+]) or die "standby2 never connected";
+
+$primary->safe_psql('postgres', q[
+    CREATE TABLE t (k text PRIMARY KEY, v int NOT NULL);
+    INSERT INTO t VALUES ('x', 0), ('y', 0);
+]);
+$primary->wait_for_catchup($standby1);
+$primary->wait_for_catchup($standby2);
+
+$standby1->safe_psql('postgres', 'SELECT pg_wal_replay_pause();');
+
+my $tx1 = $primary->background_psql('postgres');
+$tx1->query_safe(q[
+    SET application_name = 'long_fork_t1';
+    SET synchronous_commit = 'remote_apply';
+]);
+$tx1->query_until(qr/committing/, qq[
+BEGIN;
+UPDATE t SET v = 1 WHERE k = 'x';
+\\echo committing
+COMMIT;
+]);
+
+$primary->poll_query_until('postgres', q[
+    SELECT count(*) = 1
+    FROM pg_stat_activity
+    WHERE application_name = 'long_fork_t1'
+      AND wait_event_type = 'IPC'
+      AND wait_event = 'SyncRep'
+]) or die "tx1 never reached SyncRep wait";
+
+my $lsn = $primary->safe_psql('postgres',
+	"SELECT pg_current_wal_flush_lsn();");
+chomp $lsn;
+
+$standby2->poll_query_until('postgres',
+	"SELECT pg_last_wal_replay_lsn() >= '$lsn'::pg_lsn")
+  or die "standby2 never applied tx1";
+
+$standby2->safe_psql('postgres', 'SELECT pg_wal_replay_pause();');
+
+$primary->safe_psql('postgres', q[
+    SET synchronous_commit = off;
+    BEGIN;
+    UPDATE t SET v = 1 WHERE k = 'y';
+    COMMIT;
+]);
+
+my $m = $primary->safe_psql('postgres',
+	"SELECT string_agg(v::text, ',' ORDER BY k) FROM t;");
+my $s = $standby2->safe_psql('postgres',
+	"SELECT string_agg(v::text, ',' ORDER BY k) FROM t;");
+
+is($m, '0,1', 'primary sees T2 but not T1');
+is($s, '1,0', 'standby sees T1 but not T2');
+
+$standby1->safe_psql('postgres', 'SELECT pg_wal_replay_resume();');
+$standby2->safe_psql('postgres', 'SELECT pg_wal_replay_resume();');
+$tx1->quit;
+$primary->wait_for_catchup($standby1);
+$primary->wait_for_catchup($standby2);
+
+done_testing();