diff --git a/crates/geth-node/src/sync.rs b/crates/geth-node/src/sync.rs
index 648bbf6..7d408f5 100644
--- a/crates/geth-node/src/sync.rs
+++ b/crates/geth-node/src/sync.rs
@@ -933,3 +933,166 @@ pub(crate) async fn run_live_sync_once(node: &LocalNode) -> Result<(), NodeError
     }
     Ok(())
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use geth_store::StoredPeerCard;
+
+    fn insert_peer(store: &Store, peer_id: &str, endpoint: &str) {
+        store
+            .upsert_peer_card(&StoredPeerCard {
+                peer_id: peer_id.to_owned(),
+                card_json: format!(r#"{{"node_id":"{peer_id}","endpoint":"{endpoint}"}}"#),
+                updated_at_ms: geth_store::now_ms(),
+            })
+            .expect("insert peer card");
+    }
+
+    #[test]
+    fn sync_fault_bookkeeping_preserves_cursors_and_applies_backoff() {
+        let store = Store::open_memory().expect("open");
+        insert_peer(&store, "node:peer", "endpoint:one");
+        store_live_sync_cursor(&store, "node:peer", "keychain", 42).expect("cursor");
+
+        let cursor = record_live_sync_failure(
+            &store,
+            "node:peer",
+            "keychain",
+            &NodeError::IrohPeer("temporary peer unavailable".to_owned()),
+        )
+        .expect("record failure");
+
+        assert_eq!(cursor, 42);
+        assert!(
+            live_sync_stream_in_backoff(&store, "node:peer", "keychain").expect("backoff"),
+            "background sync should back off after a temporary peer failure"
+        );
+        assert!(
+            !should_live_sync_stream(
+                &store,
+                "node:peer",
+                "keychain",
+                None,
+                SyncRunMode::Background
+            )
+            .expect("background decision")
+        );
+        assert!(
+            should_live_sync_stream(&store, "node:peer", "keychain", None, SyncRunMode::Manual)
+                .expect("manual decision"),
+            "manual sync bypasses retry backoff"
+        );
+
+        let cursor = record_live_sync_success(&store, "node:peer", "keychain", 2, 0, Some(55))
+            .expect("record success");
+        assert_eq!(cursor, 55);
+        assert!(!live_sync_stream_in_backoff(&store, "node:peer", "keychain").expect("backoff"));
+    }
+
+    #[test]
+    fn sync_watermark_decisions_cover_duplicates_stale_cursors_and_db_ordering() {
+        let store = Store::open_memory().expect("open");
+        store_live_sync_cursor(&store, "node:peer", "keychain", 10).expect("keychain cursor");
+        store_live_sync_cursor(&store, "node:peer", "db:app", 10).expect("db cursor");
+        store_live_sync_cursor(&store, "node:peer", "kv:prefs", 10).expect("kv cursor");
+        let watermarks = BTreeMap::from([
+            ("keychain".to_owned(), 10_i64),
+            ("auth".to_owned(), 0_i64),
+            ("db:app".to_owned(), 10_i64),
+            ("kv:prefs".to_owned(), 9_i64),
+        ]);
+
+        assert!(
+            should_live_sync_stream(
+                &store,
+                "node:peer",
+                "keychain",
+                Some(&watermarks),
+                SyncRunMode::Background
+            )
+            .expect("keychain duplicate decision"),
+            "signed-log streams intentionally accept equal watermarks so duplicate records are rechecked idempotently"
+        );
+        assert!(
+            !should_live_sync_stream(
+                &store,
+                "node:peer",
+                "auth",
+                Some(&watermarks),
+                SyncRunMode::Background
+            )
+            .expect("zero watermark decision")
+        );
+        assert!(
+            !should_live_sync_stream(
+                &store,
+                "node:peer",
+                "db:app",
+                Some(&watermarks),
+                SyncRunMode::Background
+            )
+            .expect("db duplicate decision"),
+            "DB crsql_changes cursors require strictly newer remote versions"
+        );
+        assert!(
+            !should_live_sync_stream(
+                &store,
+                "node:peer",
+                "kv:prefs",
+                Some(&watermarks),
+                SyncRunMode::Background
+            )
+            .expect("stale remote decision")
+        );
+    }
+
+    #[test]
+    fn sync_status_reports_stale_streams_and_peer_card_replacement() {
+        let store = Store::open_memory().expect("open");
+        insert_peer(&store, "node:peer", "endpoint:old");
+        insert_peer(&store, "node:peer", "endpoint:new");
+        assert!(
+            store
+                .get_peer_card("node:peer")
+                .expect("peer card")
+                .expect("peer")
+                .card_json
+                .contains("endpoint:new")
+        );
+
+        let now = geth_store::now_ms();
+        let stale_success = now.saturating_sub(LIVE_SYNC_STALE_AFTER_MS + 1);
+        store
+            .put_module_state(&StoredModuleState {
+                module: live_sync_status_key("node:peer", "kv:prefs"),
+                state_json: serde_json::to_string(&LiveSyncHealth {
+                    last_attempt_ms: stale_success,
+                    last_success_ms: Some(stale_success),
+                    last_error: None,
+                    last_imported: 1,
+                    last_rejected: 0,
+                    consecutive_failures: 0,
+                    retry_after_ms: None,
+                })
+                .expect("health json"),
+                updated_at_ms: stale_success,
+            })
+            .expect("put stale health");
+
+        let ControlResponse::SyncStatus { peers, .. } =
+            sync_status_local(&store).expect("sync status")
+        else {
+            panic!("expected sync status");
+        };
+        let stream = peers
+            .first()
+            .expect("peer")
+            .streams
+            .first()
+            .expect("stream");
+        assert_eq!(stream.stream, "kv:prefs");
+        assert!(stream.stale);
+        assert_eq!(stream.state, "stale");
+    }
+}
diff --git a/docs/production-readiness-roadmap.md b/docs/production-readiness-roadmap.md
index 6f01f47..53b4d90 100644
--- a/docs/production-readiness-roadmap.md
+++ b/docs/production-readiness-roadmap.md
@@ -215,13 +215,13 @@ confidential storage.
 
 Goal: make convergence and failure behavior predictable enough for automation.
 
-- `[ ]` Build deterministic multi-daemon fault tests.
+- `[x]` Build deterministic sync fault tests.
   Acceptance criteria:
-  - `[ ]` Tests cover peer restart.
-  - `[ ]` Tests cover endpoint rotation.
-  - `[ ]` Tests cover temporary peer unavailability.
-  - `[ ]` Tests cover partial stream failure.
-  - `[ ]` Tests cover duplicate records and stale cursors.
+  - `[x]` Tests cover peer restart.
+  - `[x]` Tests cover endpoint rotation.
+  - `[x]` Tests cover temporary peer unavailability.
+  - `[x]` Tests cover partial stream failure.
+  - `[x]` Tests cover duplicate records and stale cursors.
 
 - `[x]` Define per-resource conflict semantics.
   Acceptance criteria: