diff --git a/scripts/pb-add-dn-staged-ingestion.py b/scripts/pb-add-dn-staged-ingestion.py
index 08a6ed4..d6e497e 100644
--- a/scripts/pb-add-dn-staged-ingestion.py
+++ b/scripts/pb-add-dn-staged-ingestion.py
@@ -57,7 +57,18 @@ def index_name(statement):
     match = re.search(r"\bINDEX\s+`?([^`\s]+)`?", statement, re.IGNORECASE)
     if not match:
         raise ValueError(f"not a CREATE INDEX statement: {statement!r}")
-    return match.group(1)
+    return match.group(1).strip().lower()
+
+
+def normalise_index_sql(statement):
+    """Compare index SQL by meaning, not PocketBase's formatting choices."""
+    value = statement.strip().rstrip(";")
+    value = re.sub(r"`([^`]*)`", lambda match: f"`{match.group(1).lower()}`", value)
+    value = re.sub(r"\s+", " ", value).strip().lower()
+    # SQLite/PocketBase may insert or remove spaces around these tokens while
+    # preserving the same columns, uniqueness and WHERE predicate.
+    value = re.sub(r"\s*([(),=<>!])\s*", r"\1", value)
+    return value
 
 
 def merge_indexes(existing, wanted):
@@ -72,7 +83,7 @@ def merge_indexes(existing, wanted):
             positions[name] = len(merged)
             merged.append(statement)
             changes.append(("add", statement))
-        elif merged[position] != statement:
+        elif normalise_index_sql(merged[position]) != normalise_index_sql(statement):
             merged[position] = statement
             changes.append(("replace", statement))
     return merged, changes
@@ -177,6 +188,26 @@ def self_test():
     assert candidate_indexes == [candidate_index]
     assert candidate_changes == [("add", candidate_index)]
     assert merge_indexes(candidate_indexes, [candidate_index])[1] == []
+    # PocketBase returns the same three partial indexes uppercased and with
+    # multiline/variable whitespace. Those rows must be idempotent.
+    semantic_wanted = [
+        "CREATE UNIQUE INDEX `idx_dn_batch_notes_candidate_key` ON `dn_batch_notes` (`batch`, `candidate_key`) WHERE `candidate_key` IS NOT NULL AND `candidate_key` != ''",
+        "CREATE UNIQUE INDEX `idx_delivery_notes_batch_note` ON `delivery_notes` (`batch_note`) WHERE `batch_note` IS NOT NULL AND `batch_note` != ''",
+        "CREATE UNIQUE INDEX `idx_dn_batches_capture_session_id` ON `dn_batches` (`capture_session_id`) WHERE `capture_session_id` != ''",
+    ]
+    pocketbase_stored = [
+        "CREATE UNIQUE INDEX `IDX_DN_BATCH_NOTES_CANDIDATE_KEY`\nON `DN_BATCH_NOTES` (`BATCH`,  `CANDIDATE_KEY`)\nWHERE `CANDIDATE_KEY` IS NOT NULL AND `CANDIDATE_KEY` != ''",
+        "CREATE UNIQUE INDEX `IDX_DELIVERY_NOTES_BATCH_NOTE` ON `DELIVERY_NOTES`\n( `BATCH_NOTE` ) WHERE `BATCH_NOTE` IS NOT NULL\nAND `BATCH_NOTE` != ''",
+        "CREATE UNIQUE INDEX `IDX_DN_BATCHES_CAPTURE_SESSION_ID` ON `DN_BATCHES` (`CAPTURE_SESSION_ID`)\nWHERE `CAPTURE_SESSION_ID` != ''",
+    ]
+    assert merge_indexes(pocketbase_stored, semantic_wanted)[1] == []
+    near_misses = [
+        semantic_wanted[0].replace("candidate_key`)", "other_key`)") ,
+        semantic_wanted[1].replace("CREATE UNIQUE", "CREATE"),
+        semantic_wanted[2].replace(" != ''", " IS NOT NULL"),
+    ]
+    for stored, near_miss in zip(pocketbase_stored, near_misses):
+        assert merge_indexes([stored], [near_miss])[1] == [("replace", near_miss)]
     print("self-test: index replacement and idempotence passed")
     return 0