diff --git a/src/live/finalization.rs b/src/live/finalization.rs
index 1def50e..52d4cd0 100644
--- a/src/live/finalization.rs
+++ b/src/live/finalization.rs
@@ -5,13 +5,18 @@ use crate::beacon_client::types::FinalizedCheckpointEvent;
 use crate::config::EffectiveScanMode;
 use crate::db::Pool as PgPool;
 use crate::db::scanner as db_scanner;
-use crate::error::Result;
+use crate::error::{Error, Result};
 use crate::exits;
 use crate::scanner;
 
+fn is_beacon_request_failure(e: &Error) -> bool {
+    matches!(e, Error::Http(_) | Error::BeaconApi { .. })
+}
+
 #[allow(clippy::too_many_arguments)]
 pub(super) async fn process_finalized_rescan(
     client: &BeaconClient,
+    backfill_client: Option<&BeaconClient>,
     pool: &PgPool,
     scan_validators: &HashSet<u64>,
     validator_exits: &HashMap<u64, u64>,
@@ -47,12 +52,32 @@ pub(super) async fn process_finalized_rescan(
         if active.is_empty() {
             continue;
         }
-        scanner::scan_epoch(client, pool, epoch, &active, true, scan_mode)
-            .await
-            .map_err(|e| {
+        match scanner::scan_epoch(client, pool, epoch, &active, true, scan_mode).await {
+            Ok(()) => {}
+            Err(e) if is_beacon_request_failure(&e) && backfill_client.is_some() => {
+                let bf = backfill_client.expect("checked above");
+                tracing::warn!(
+                    epoch,
+                    error = %e,
+                    "Live client failed re-scan; retrying with backfill client"
+                );
+                scanner::scan_epoch(bf, pool, epoch, &active, true, scan_mode)
+                    .await
+                    .map_err(|e2| {
+                        tracing::error!(
+                            epoch,
+                            primary_error = %e,
+                            backfill_error = %e2,
+                            "Backfill client also failed re-scan; aborting"
+                        );
+                        e2
+                    })?;
+            }
+            Err(e) => {
                 tracing::error!(epoch, error = %e, "Failed to re-scan finalized epoch; aborting");
-                e
-            })?;
+                return Err(e);
+            }
+        }
     }
     crate::metrics::LIVE_FINALIZED_RESCAN_DURATION
         .with_label_values(&["rescan_loop"])
diff --git a/src/live/mod.rs b/src/live/mod.rs
index 43af2bd..015eb68 100644
--- a/src/live/mod.rs
+++ b/src/live/mod.rs
@@ -25,6 +25,7 @@ use reorg::process_chain_reorg;
 #[allow(clippy::too_many_arguments)]
 pub async fn run_live_tracking(
     client: &BeaconClient,
+    backfill_client: Option<&BeaconClient>,
     pool: &PgPool,
     instance_id: Uuid,
     tracked: &HashSet<u64>,
@@ -114,6 +115,7 @@ pub async fn run_live_tracking(
                                         .inc();
                                     process_finalized_rescan(
                                         client,
+                                        backfill_client,
                                         pool,
                                         &scan_validators,
                                         validator_exits,
diff --git a/src/main.rs b/src/main.rs
index cb278ef..dbf7d93 100644
--- a/src/main.rs
+++ b/src/main.rs
@@ -310,8 +310,16 @@ async fn main() -> anyhow::Result<()> {
     } else {
         // Concurrent mode: live in the foreground, backfill in the background.
         // SSE events are never blocked by the slower historical scan.
+        // Only pass a backfill client to the live loop when it's a *separate*
+        // beacon node — otherwise it's just an Arc clone of the live client
+        // and retrying against it on a finalized-rescan failure is wasted.
+        let live_backfill = config
+            .backfill_beacon_url
+            .as_ref()
+            .map(|_| backfill_client.as_ref());
         let live_result = live::run_live_tracking(
             live_client.as_ref(),
+            live_backfill,
             &pool,
             instance_id,
             &tracked_set,