diff --git a/apps/spark/src/main/java/com/linkedin/openhouse/jobs/spark/Operations.java b/apps/spark/src/main/java/com/linkedin/openhouse/jobs/spark/Operations.java
index bd301c729..b9feb9dfb 100644
--- a/apps/spark/src/main/java/com/linkedin/openhouse/jobs/spark/Operations.java
+++ b/apps/spark/src/main/java/com/linkedin/openhouse/jobs/spark/Operations.java
@@ -14,7 +14,6 @@
 import com.linkedin.openhouse.jobs.util.SparkJobUtil;
 import com.linkedin.openhouse.jobs.util.TableStatsCollector;
 import java.io.IOException;
-import java.nio.file.Paths;
 import java.time.ZonedDateTime;
 import java.time.temporal.ChronoUnit;
 import java.util.Collections;
@@ -356,7 +355,7 @@ private Map<String, List<String>> prepareBackupDataManifests(
               Collectors.groupingBy(
                   task -> {
                     String path = task.file().path().toString();
-                    return Paths.get(path).getParent().toString();
+                    return dataFileParent(path);
                   },
                   Collectors.mapping(task -> task.file().path().toString(), Collectors.toList())));
     } catch (IOException e) {
@@ -403,7 +402,20 @@ private void exposeBackupLocation(Table table, String backupDir) {
             table.name(), AppConstants.BACKUP_DIR_KEY, fullyQualifiedBackupDir));
   }
 
-  private Path getTrashPath(String path, String filePath, String trashDir) {
+  /**
+   * Return the parent directory of a data file path. Uses Hadoop {@link Path} (not {@link
+   * java.nio.file.Paths}, which is not URI-aware) so that the scheme and authority of a
+   * fully-qualified path such as {@code hdfs://cluster/a/b/f.orc} are preserved. Dropping the
+   * authority here would later resolve the backup manifest against the default filesystem root and
+   * fail with a permission error.
+   */
+  @VisibleForTesting
+  static String dataFileParent(String dataFilePath) {
+    return new Path(dataFilePath).getParent().toString();
+  }
+
+  @VisibleForTesting
+  static Path getTrashPath(String path, String filePath, String trashDir) {
     return new Path(filePath.replace(path, new Path(path, trashDir).toString()));
   }
 
diff --git a/apps/spark/src/test/java/com/linkedin/openhouse/jobs/spark/OperationsBackupPathTest.java b/apps/spark/src/test/java/com/linkedin/openhouse/jobs/spark/OperationsBackupPathTest.java
new file mode 100644
index 000000000..0fd3a83c2
--- /dev/null
+++ b/apps/spark/src/test/java/com/linkedin/openhouse/jobs/spark/OperationsBackupPathTest.java
@@ -0,0 +1,72 @@
+package com.linkedin.openhouse.jobs.spark;
+
+import org.apache.hadoop.fs.Path;
+import org.junit.jupiter.api.Assertions;
+import org.junit.jupiter.api.Test;
+
+/**
+ * Pure-Java unit tests for the backup-manifest path derivation in {@link Operations}. No Spark
+ * session, no filesystem — exercises the scheme/authority handling that broke retention backups
+ * when data files are stored as fully-qualified {@code hdfs://authority/...} URIs.
+ *
+ * <p>These guard against a regression where {@code java.nio.file.Paths} was used to derive parent
+ * paths: it is not URI-aware and collapses {@code hdfs://authority} into {@code hdfs:/authority},
+ * dropping the authority. The mangled path later resolves against the default filesystem root and
+ * fails with {@code Permission denied ... inode="/"}.
+ */
+public class OperationsBackupPathTest {
+
+  @Test
+  public void dataFileParentPreservesSchemeAndAuthority() {
+    String dataFile =
+        "hdfs://ltx1-holdem/data/openhouse/db/tbl-uuid/data/datepartition=2026-03-26-00/f.orc";
+
+    String parent = Operations.dataFileParent(dataFile);
+
+    // Authority (//ltx1-holdem) must be retained; the old Paths.get path produced
+    // "hdfs:/ltx1-holdem/..." (single slash, authority dropped).
+    Assertions.assertEquals(
+        "hdfs://ltx1-holdem/data/openhouse/db/tbl-uuid/data/datepartition=2026-03-26-00", parent);
+  }
+
+  @Test
+  public void dataFileParentPreservesBarePath() {
+    String dataFile = "/data/openhouse/db/tbl-uuid/data/datepartition=2026-03-06-00/f.orc";
+
+    Assertions.assertEquals(
+        "/data/openhouse/db/tbl-uuid/data/datepartition=2026-03-06-00",
+        Operations.dataFileParent(dataFile));
+  }
+
+  @Test
+  public void manifestDestPathForQualifiedDataFileKeepsAuthorityAndInsertsBackupDir() {
+    // Mirrors the production chain in prepareBackupDataManifests + writeBackupDataManifests.
+    // Iceberg stores table.location() bare (no scheme), but newer data files are fully qualified.
+    String tableLocation = "/data/openhouse/db/tbl-uuid";
+    String dataFile =
+        "hdfs://ltx1-holdem/data/openhouse/db/tbl-uuid/data/datepartition=2026-03-26-00/f.orc";
+
+    String partitionPath = Operations.dataFileParent(dataFile);
+    String manifestSource = new Path(partitionPath, "data_manifest_1.json").toString();
+    Path dest = Operations.getTrashPath(tableLocation, manifestSource, ".backup");
+
+    Assertions.assertEquals(
+        "hdfs://ltx1-holdem/data/openhouse/db/tbl-uuid/.backup/data/"
+            + "datepartition=2026-03-26-00/data_manifest_1.json",
+        dest.toString());
+  }
+
+  @Test
+  public void manifestDestPathForBareDataFileInsertsBackupDir() {
+    String tableLocation = "/data/openhouse/db/tbl-uuid";
+    String dataFile = "/data/openhouse/db/tbl-uuid/data/datepartition=2026-03-06-00/f.orc";
+
+    String partitionPath = Operations.dataFileParent(dataFile);
+    String manifestSource = new Path(partitionPath, "data_manifest_1.json").toString();
+    Path dest = Operations.getTrashPath(tableLocation, manifestSource, ".backup");
+
+    Assertions.assertEquals(
+        "/data/openhouse/db/tbl-uuid/.backup/data/datepartition=2026-03-06-00/data_manifest_1.json",
+        dest.toString());
+  }
+}