diff --git a/pkg/connector/backfill.go b/pkg/connector/backfill.go
index 6a6a4255..942024d0 100644
--- a/pkg/connector/backfill.go
+++ b/pkg/connector/backfill.go
@@ -321,7 +321,29 @@ func (m *MetaClient) wrapBackfillEvents(ctx context.Context, portal *bridgev2.Po
 	if anchor != nil {
 		if forward {
 			upsert.Messages = slices.DeleteFunc(upsert.Messages, func(message *table.WrappedMessage) bool {
-				return message.TimestampMs <= anchor.Timestamp.UnixMilli()
+				if message.TimestampMs > anchor.Timestamp.UnixMilli() {
+					return false
+				}
+				// This message is older than the portal's latest bridged message, so it would
+				// normally be assumed to already exist. However, Meta can skip threads in a
+				// reconnect snapshot while advancing the sync cursor past their messages (see
+				// PLAT-36990), which leaves permanent mid-timeline gaps. Only drop the message
+				// if it actually exists in the bridge database, otherwise recover it.
+				existing, err := m.Main.Bridge.DB.Message.GetFirstPartByID(ctx, m.UserLogin.ID, metaid.MakeFBMessageID(message.MessageId))
+				if err != nil {
+					zerolog.Ctx(ctx).Err(err).
+						Str("message_id", message.MessageId).
+						Msg("Failed to check message existence during forward backfill dedup, dropping")
+					return true
+				}
+				if existing == nil {
+					zerolog.Ctx(ctx).Warn().
+						Str("message_id", message.MessageId).
+						Int64("timestamp_ms", message.TimestampMs).
+						Msg("Recovering pre-anchor message missing from database in forward backfill")
+					return false
+				}
+				return true
 			})
 		} else {
 			upsert.Messages = slices.DeleteFunc(upsert.Messages, func(message *table.WrappedMessage) bool {
diff --git a/pkg/messagix/syncManager.go b/pkg/messagix/syncManager.go
index e48ec4ef..428e8890 100644
--- a/pkg/messagix/syncManager.go
+++ b/pkg/messagix/syncManager.go
@@ -125,8 +125,12 @@ func (sm *SyncManager) SyncSocketData(ctx context.Context, databaseID int64, db
 	resp.Finish()
 
 	if len(resp.Table.LSHandleSyncFailure) > 0 {
-		// TODO handle these somehow?
-		sm.client.Logger.Warn().
+		// These failures mean the server-side sync for the database errored and no
+		// cursor was obtained, so the database stays unsynced until the next attempt.
+		// A common cause is empty sync params (handled in getSyncParams, see
+		// PLAT-36990); log at error level with the database ID so it's visible.
+		sm.client.Logger.Error().
+			Int64("database_id", databaseID).
 			Any("sync_failures", resp.Table.LSHandleSyncFailure).
 			Msg("Sync failures found")
 	}
@@ -236,18 +240,30 @@ func (sm *SyncManager) UpdateDatabaseSyncParams(dbs []*socket.QueryMetadata) err
 
 var dbID7Params = `{"mnet_rank_types":[44]}`
 
+// defaultLocaleSyncParams is the fallback sync params payload used when the page
+// config doesn't provide LSPlatformMessengerSyncParams. messenger-lite sessions
+// don't include these, so an empty string would make Meta's server fail with
+// "Invalid argument supplied for foreach()" (see PLAT-36990), permanently
+// breaking the affected sync database (e.g. Contact sync on database 2).
+var defaultLocaleSyncParams = `{"locale":"en_US"}`
+
 func (sm *SyncManager) getSyncParams(dbID int64, ch socket.SyncChannel) *string {
 	if dbID == 7 {
 		return &dbID7Params
 	}
+	var params *string
 	switch ch {
 	case socket.MailBox:
-		return &sm.syncParams.Mailbox
+		params = &sm.syncParams.Mailbox
 	case socket.Contact:
-		return &sm.syncParams.Contact
+		params = &sm.syncParams.Contact
 	default:
-		return &sm.syncParams.E2Ee
+		params = &sm.syncParams.E2Ee
+	}
+	if params == nil || *params == "" {
+		return &defaultLocaleSyncParams
 	}
+	return params
 }
 
 func (sm *SyncManager) GetCursor(db int64) string {