diff --git a/alembic/versions/2026_07_14_0900_browse_covering_index.py b/alembic/versions/2026_07_14_0900_browse_covering_index.py
new file mode 100644
index 00000000..758dba56
--- /dev/null
+++ b/alembic/versions/2026_07_14_0900_browse_covering_index.py
@@ -0,0 +1,27 @@
+"""Covering index for browse pages.
+
+The videos/files browse queries filter on mimetype and order by effective_datetime; without a
+covering index SQLite scans every file_group row (which carries the indexed texts inline), which
+takes ~18 seconds cold on a Pi.  With (mimetype, effective_datetime) the filter, count, and sort
+resolve index-only.
+
+Revision ID: 2026_07_14_0900
+Revises: 2026_07_10_1350
+"""
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision = '2026_07_14_0900'
+down_revision = '2026_07_10_1350'
+branch_labels = None
+depends_on = None
+
+
+def upgrade():
+    # IF NOT EXISTS: QA boxes may have had the index created manually before this migration.
+    op.execute('CREATE INDEX IF NOT EXISTS file_group_mimetype_effective_idx '
+               'ON file_group (mimetype, effective_datetime)')
+
+
+def downgrade():
+    op.execute('DROP INDEX IF EXISTS file_group_mimetype_effective_idx')
diff --git a/modules/archive/lib.py b/modules/archive/lib.py
index 7246aaf0..e1422c09 100644
--- a/modules/archive/lib.py
+++ b/modules/archive/lib.py
@@ -1376,8 +1376,8 @@ ARCHIVE_ORDERS = {
     '-download_datetime': 'fg.download_datetime DESC NULLS LAST',
     'rank': f'2 DESC, {_date}',
     '-rank': f'2 ASC, {date}',
-    'size': 'fg.size ASC, LOWER(fg.primary_path) ASC',
-    '-size': 'fg.size DESC NULLS LAST, LOWER(fg.primary_path) DESC',
+    'size': 'fg.size ASC, fg.id ASC',
+    '-size': 'fg.size DESC NULLS LAST, fg.id DESC',
     'viewed': 'fg.viewed ASC',
     '-viewed': 'fg.viewed DESC',
 }
@@ -1444,15 +1444,24 @@ def search_archives(search_str: str, domain: str, limit: int, offset: int, order
     select_columns = f"{select_columns}," if select_columns else ""
     wheres = '\n AND '.join(wheres)
     where = f'WHERE\n{wheres}' if wheres else ''
+    if where or fts_join:
+        from_clause = f'''FROM archive a
+            LEFT JOIN file_group fg ON fg.id = a.file_group_id
+            {fts_join}'''
+    else:
+        # Browse (no filters): CROSS JOIN pins file_group as the outer table so SQLite walks its
+        # date index and probes archive's covering index -- both sides stay index-only.  Driving
+        # from archive probes every (large) file_group row for the sort key, which takes ~90
+        # seconds cold on a Pi with 124k archives.
+        from_clause = 'FROM file_group fg CROSS JOIN archive a ON a.file_group_id = fg.id'
+
     stmt = f'''
             SELECT
                 a.file_group_id AS id, -- always get `file_group.id` for `handle_file_group_search_results`
                 {select_columns}
                 COUNT(*) OVER() AS total
                 {headline_columns}
-            FROM archive a
-            LEFT JOIN file_group fg ON fg.id = a.file_group_id
-            {fts_join}
+            {from_clause}
             {where}
             ORDER BY {order_by}
             LIMIT :limit OFFSET :offset
diff --git a/modules/videos/test/test_api.py b/modules/videos/test/test_api.py
index fe71b97c..fc3b8284 100644
--- a/modules/videos/test/test_api.py
+++ b/modules/videos/test/test_api.py
@@ -257,9 +257,10 @@ async def test_search_videos(test_session, video_factory, assert_video_search, s
                               offset=1)
     await assert_video_search(assert_total=1, assert_ids=[vid2.id], order_by='-published_datetime',
                               channel_id=simple_channel.id)
-    await assert_video_search(assert_total=2, assert_ids=[vid1.id, vid2.id], tag_names=[tag1.name])
+    # No search_str, so the default 'rank' order is a constant; ties fall to fg.id DESC.
+    await assert_video_search(assert_total=2, assert_ids=[vid2.id, vid1.id], tag_names=[tag1.name])
     await assert_video_search(assert_total=2, assert_ids=[vid3.id, vid2.id], tag_names=[tag2.name])
-    await assert_video_search(assert_total=2, assert_ids=[vid1.id], tag_names=[tag1.name], limit=1)
+    await assert_video_search(assert_total=2, assert_ids=[vid2.id], tag_names=[tag1.name], limit=1)
     # Only vid2 has both tags.
     await assert_video_search(assert_total=1, assert_ids=[vid2.id], tag_names=[tag1.name, tag2.name])
 
diff --git a/modules/videos/video/lib.py b/modules/videos/video/lib.py
index ac87cfec..b497af9b 100644
--- a/modules/videos/video/lib.py
+++ b/modules/videos/video/lib.py
@@ -50,23 +50,25 @@ def get_video(file_group_id: int) -> Video:
 
 
 VIDEO_ORDERS = {
+    # fg.id (the rowid) is the pagination tiebreaker: it is present in every index, so these
+    # orders stay index-only.  LOWER(fg.primary_path) forced SQLite to read every (large) row.
     # Sometimes we don't have a published_datetime.  This is equivalent to COALESCE(fg.published_datetime, fg.download_datetime)
-    'published_datetime': 'fg.effective_datetime ASC, LOWER(fg.primary_path) ASC',
-    '-published_datetime': 'fg.effective_datetime DESC NULLS LAST, LOWER(fg.primary_path) ASC',
-    'rank': '2 DESC, LOWER(fg.primary_path) DESC',
-    '-rank': '2 ASC, LOWER(fg.primary_path) ASC',
-    'size': 'fg.size ASC, LOWER(fg.primary_path) ASC',
-    '-size': 'fg.size DESC, LOWER(fg.primary_path) DESC',
-    'length': 'fg.length ASC, LOWER(fg.primary_path) ASC',
-    '-length': 'fg.length DESC, LOWER(fg.primary_path) DESC',
-    'size_to_duration': 'CASE WHEN fg.length > 0 THEN CAST(fg.size AS REAL) / fg.length ELSE fg.size END ASC, LOWER(fg.primary_path) ASC',
-    '-size_to_duration': 'CASE WHEN fg.length > 0 THEN CAST(fg.size AS REAL) / fg.length ELSE fg.size END DESC, LOWER(fg.primary_path) DESC',
+    'published_datetime': 'fg.effective_datetime ASC, fg.id ASC',
+    '-published_datetime': 'fg.effective_datetime DESC NULLS LAST, fg.id ASC',
+    'rank': '2 DESC, fg.id DESC',
+    '-rank': '2 ASC, fg.id ASC',
+    'size': 'fg.size ASC, fg.id ASC',
+    '-size': 'fg.size DESC, fg.id DESC',
+    'length': 'fg.length ASC, fg.id ASC',
+    '-length': 'fg.length DESC, fg.id DESC',
+    'size_to_duration': 'CASE WHEN fg.length > 0 THEN CAST(fg.size AS REAL) / fg.length ELSE fg.size END ASC, fg.id ASC',
+    '-size_to_duration': 'CASE WHEN fg.length > 0 THEN CAST(fg.size AS REAL) / fg.length ELSE fg.size END DESC, fg.id DESC',
     'viewed': 'fg.viewed ASC',
     '-viewed': 'fg.viewed DESC',
     'view_count': 'v.view_count ASC',
     '-view_count': 'v.view_count DESC',
-    'download_datetime': 'fg.download_datetime ASC, LOWER(fg.primary_path) ASC',
-    '-download_datetime': 'fg.download_datetime DESC NULLS LAST, LOWER(fg.primary_path) ASC',
+    'download_datetime': 'fg.download_datetime ASC, fg.id ASC',
+    '-download_datetime': 'fg.download_datetime DESC NULLS LAST, fg.id ASC',
 }
 NO_NULL_ORDERS = {
     'viewed': 'fg.viewed IS NOT NULL',
diff --git a/wrolpi/files/models.py b/wrolpi/files/models.py
index 1d37f436..b5656f88 100644
--- a/wrolpi/files/models.py
+++ b/wrolpi/files/models.py
@@ -72,6 +72,9 @@ class FileGroup(ModelHelper, Base):
         Index('file_group_download_datetime_idx', 'download_datetime'),
         Index('file_group_effective_datetime_idx', 'effective_datetime'),
         Index('file_group_length_idx', 'length'),
+        # Covering index for browse pages: mimetype filter + date order resolve index-only
+        # instead of scanning every (large) file_group row.
+        Index('file_group_mimetype_effective_idx', 'mimetype', 'effective_datetime'),
         Index('file_group_mimetype_idx', 'mimetype'),
         Index('file_group_model_idx', 'model'),
         Index('file_group_modification_datetime_idx', 'modification_datetime'),
diff --git a/wrolpi/files/worker.py b/wrolpi/files/worker.py
index 445f7eee..88e9c32c 100644
--- a/wrolpi/files/worker.py
+++ b/wrolpi/files/worker.py
@@ -1269,6 +1269,12 @@ class FileWorker:
 
         # Set refresh_complete flag only when refreshing the entire media directory
         if is_global_refresh:
+            # Refresh statistics for the query planner; without sqlite_stat1 (or after the table
+            # contents change wholesale) SQLite picks full scans over the covering indexes.
+            from wrolpi.db import get_db_curs
+            with get_db_curs(commit=True) as curs:
+                curs.execute('PRAGMA optimize')
+
             flags.refresh_complete.set()
             flags.global_refresh_active.clear()
             # After a DB rebuild, playlists.yaml was imported before any files were indexed, so its