diff --git a/benchmark/result_consumers_bench.dart b/benchmark/result_consumers_bench.dart
new file mode 100644
index 00000000..9873df2f
--- /dev/null
+++ b/benchmark/result_consumers_bench.dart
@@ -0,0 +1,78 @@
+// What the consumers of a big result set cost through the dataset's lazy row
+// list versus a `List<List<String>>` built from it (#1549). Headless:
+//   flutter test benchmark/result_consumers_bench.dart --dart-define=ROWS=50000
+// Prints `RESULT <consumer> materialized=<ms> lazy=<ms>` (median of 5, after a
+// warm-up). Memory is not measured here: RSS in the test runner is too noisy.
+// The 50k x 20 figures in #1549 (68.6 MiB as strings, 17.5 MiB columnar) came
+// from a standalone AOT run.
+//
+// Decision (see #1549): consumers keep `List<List<String>>` at their boundary.
+// At 50k x 20 the lazy list is as fast for sort and filter on typed columns,
+// and 25-40 ms slower for a free-text filter and the CSV export, which run
+// once per user action; the app only ever holds the columnar dataset.
+import 'package:flutter_test/flutter_test.dart';
+import 'package:querya_desktop/core/database/compact_result_dataset.dart';
+import 'package:querya_desktop/features/workspace/grid/grid_sorting.dart';
+import 'package:querya_desktop/features/workspace/grid_filter_engine.dart';
+import 'package:querya_desktop/shared/services/data_export_service.dart';
+
+const _rows = int.fromEnvironment('ROWS', defaultValue: 50000);
+
+List<List<Object?>> _source(int n) => [
+      for (var i = 0; i < n; i++)
+        [
+          for (var c = 0; c < 20; c++)
+            c < 6
+                ? i * 7 % 10007 + c
+                : c < 10
+                    ? i * 0.37 + c
+                    : 'v${(i * 31) % 5000}_$c',
+        ],
+    ];
+
+void main() {
+  test('result set consumers: lazy dataset rows vs materialized strings', () {
+    final names = [for (var c = 0; c < 20; c++) 'c$c'];
+
+    final dataset = CompactResultDataset.fromRawRows(names, _source(_rows));
+    final materialized = [for (var i = 0; i < _rows; i++) dataset.rowStrings(i)];
+    final lazy = dataset.asLazyRowList();
+    int time(void Function() f) {
+      final sw = Stopwatch()..start();
+      f();
+      return sw.elapsedMilliseconds;
+    }
+
+    void consumer(String name, void Function(List<List<String>>) run) {
+      run(materialized);
+      run(lazy);
+      final m = [for (var i = 0; i < 5; i++) time(() => run(materialized))]
+        ..sort();
+      final l = [for (var i = 0; i < 5; i++) time(() => run(lazy))]..sort();
+      // ignore: avoid_print
+      print('RESULT $name materialized=${m[2]}ms lazy=${l[2]}ms');
+    }
+
+    consumer(
+        'sort int column',
+        (r) => sortResultGridRowsWithIndices(
+            rows: r, columnIndex: 0, order: ResultGridSortOrder.ascending));
+    consumer(
+        'sort double column',
+        (r) => sortResultGridRowsWithIndices(
+            rows: r, columnIndex: 7, order: ResultGridSortOrder.ascending));
+    consumer(
+        'sort string column',
+        (r) => sortResultGridRowsWithIndices(
+            rows: r, columnIndex: 12, order: ResultGridSortOrder.descending));
+    consumer(
+        'filter c0 > 5000',
+        (r) => GridFilterEngine.filterRowIndices(
+            filterText: 'c0 > 5000', columns: names, rows: r));
+    consumer(
+        'filter free text',
+        (r) => GridFilterEngine.filterRowIndices(
+            filterText: 'v42', columns: names, rows: r));
+    consumer('csv export', (r) => DataExportService.formatCsv(names, r));
+  });
+}