diff --git a/internal/application/service/wiki_ingest.go b/internal/application/service/wiki_ingest.go
index 603c77b65..984bc4381 100644
--- a/internal/application/service/wiki_ingest.go
+++ b/internal/application/service/wiki_ingest.go
@@ -322,7 +322,7 @@ func (s *wikiIngestService) peekPendingList(ctx context.Context, kbID string) ([
 		unique = append(unique, reversedUnique[i])
 	}
 
-	return unique, len(ops)
+	return unique, len(result)
 }
 
 // trimPendingList removes the first `count` items from the Redis pending list.
@@ -336,6 +336,28 @@ func (s *wikiIngestService) trimPendingList(ctx context.Context, kbID string, co
 	}
 }
 
+// requeueFailedOps appends failed operations back to the pending list so they
+// are retried in the next follow-up batch. Called after trimPendingList has
+// already removed the consumed batch head.
+func (s *wikiIngestService) requeueFailedOps(ctx context.Context, kbID string, ops []WikiPendingOp) {
+	if s.redisClient == nil {
+		return
+	}
+	pendingKey := wikiPendingKeyPrefix + kbID
+	for _, op := range ops {
+		data, err := json.Marshal(op)
+		if err != nil {
+			logger.Warnf(ctx, "wiki ingest: failed to marshal op for requeue: %v", err)
+			continue
+		}
+		if err := s.redisClient.RPush(ctx, pendingKey, string(data)).Err(); err != nil {
+			logger.Warnf(ctx, "wiki ingest: failed to requeue op %s: %v", op.KnowledgeID, err)
+			continue
+		}
+		logger.Infof(ctx, "wiki ingest: re-queued failed op %s (%s) for retry", op.KnowledgeID, op.DocTitle)
+	}
+}
+
 // docIngestResult captures per-document info for batch post-processing.
 type docIngestResult struct {
 	KnowledgeID string
diff --git a/internal/application/service/wiki_ingest_batch.go b/internal/application/service/wiki_ingest_batch.go
index c07d7736b..638a0aa2b 100644
--- a/internal/application/service/wiki_ingest_batch.go
+++ b/internal/application/service/wiki_ingest_batch.go
@@ -228,6 +228,7 @@ func (s *wikiIngestService) ProcessWikiIngest(ctx context.Context, t *asynq.Task
 
 	// 1. MAP PHASE (Parallel extraction and generation of updates)
 	var mapMu sync.Mutex
+	var failedOps []WikiPendingOp
 	slugUpdates := make(map[string][]SlugUpdate)
 	var docResults []*docIngestResult
 	var retractChangeDesc strings.Builder
@@ -310,6 +311,7 @@ func (s *wikiIngestService) ProcessWikiIngest(ctx context.Context, t *asynq.Task
 			if err != nil {
 				mapMu.Lock()
 				ingestFailed++
+				failedOps = append(failedOps, op)
 				mapMu.Unlock()
 				logger.Warnf(mapCtx, "wiki ingest: failed to map knowledge %s: %v", op.KnowledgeID, err)
 				return nil // Don't fail the whole batch
@@ -413,6 +415,13 @@ func (s *wikiIngestService) ProcessWikiIngest(ctx context.Context, t *asynq.Task
 
 	s.trimPendingList(ctx, payload.KnowledgeBaseID, peekedCount)
 
+	// Re-enqueue failed ops so they get retried in the next follow-up batch.
+	// This must happen after trim: trim removes the consumed batch head, then
+	// requeue appends the failed items to the tail for a future attempt.
+	if len(failedOps) > 0 {
+		s.requeueFailedOps(ctx, payload.KnowledgeBaseID, failedOps)
+	}
+
 	logger.Infof(ctx, "wiki ingest: batch completed for KB %s, %d ops, %d pages affected", payload.KnowledgeBaseID, len(pendingOps), len(allPagesAffected))
 
 	followUpScheduled = s.scheduleFollowUp(ctx, payload)