ci: fix benchmark gate flakiness on ubuntu-latest

Author: clean6378-max-it

benchmarks/baselines.json

@@ -1,32 +1,32 @@
 {
-  "_note": "Gated means from max of two ubuntu-latest CI runs (slow/1.19 margin). Sub-100us cache lookup benches excluded from gate due to runner variance.",
-  "updated": "2026-07-15T16:14:31Z",
+  "_note": "Gated means from max of three ubuntu-latest CI runs, divided by 1.19 (targets ~1.19x on slowest runner). Sub-100us cache lookup benches excluded from gate.",
+  "updated": "2026-07-15T16:40:00Z",
   "machine": "Linux",
   "groups": {
-    "parse": {
-      "test_list_workspace_projects_nocache[composers-10]": 0.009873695869187243,
-      "test_list_workspace_projects_nocache[composers-200]": 0.1458800642857019,
-      "test_list_workspace_projects_nocache[composers-50]": 0.04375626806723154
-    },
     "export": {
-      "test_post_export_zip[composers-10]": 0.006389603808617876,
-      "test_post_export_zip[composers-50]": 0.02472151171218505
+      "test_post_export_zip[composers-10]": 0.006890030582232384,
+      "test_post_export_zip[composers-50]": 0.02651060126050394
+    },
+    "parse": {
+      "test_list_workspace_projects_nocache[composers-10]": 0.010318600237016761,
+      "test_list_workspace_projects_nocache[composers-200]": 0.15541806190476232,
+      "test_list_workspace_projects_nocache[composers-50]": 0.04608288053221206
     },
     "search": {
-      "test_search_full_corpus_indexed": 0.03357773425985082,
-      "test_search_full_corpus_live_scan": 0.02699565326798046
+      "test_search_full_corpus_indexed": 0.035584541499678296,
+      "test_search_full_corpus_live_scan": 0.02765475991102243
     },
     "summary-cache": {
-      "test_composer_map_cache_lookup[hit]": 9.159433254539677e-05,
-      "test_composer_map_cache_lookup[miss]": 9.052273017297571e-05,
-      "test_fingerprint_workspace_entries[10]": 0.0014558767614049227,
-      "test_fingerprint_workspace_entries[200]": 0.01772436102941095,
-      "test_fingerprint_workspace_entries[50]": 0.006177960517756895,
-      "test_summary_cache_lookup[hit]": 9.256659141385228e-05,
-      "test_summary_cache_lookup[miss]": 9.249739842055407e-05,
-      "test_summary_cache_round_trip": 0.00051773609280467,
-      "test_tab_summary_cache_lookup[hit]": 0.00010446294267543434,
-      "test_tab_summary_cache_lookup[miss]": 0.00010424118202523503
+      "test_composer_map_cache_lookup[hit]": 5.162541185094655e-05,
+      "test_composer_map_cache_lookup[miss]": 5.0766855798028954e-05,
+      "test_fingerprint_workspace_entries[10]": 0.0015340259698651347,
+      "test_fingerprint_workspace_entries[200]": 0.019140506097323155,
+      "test_fingerprint_workspace_entries[50]": 0.006636160535039493,
+      "test_summary_cache_lookup[hit]": 5.213728423690053e-05,
+      "test_summary_cache_lookup[miss]": 5.181927082383982e-05,
+      "test_summary_cache_round_trip": 0.0002906505793927613,
+      "test_tab_summary_cache_lookup[hit]": 5.963107194614513e-05,
+      "test_tab_summary_cache_lookup[miss]": 5.868233534912213e-05
     }
   }
 }

scripts/check_benchmark_regression.py

@@ -17,11 +17,11 @@
 EXCLUDED_FROM_GATE: frozenset[str] = frozenset(
     {
         # round_trip calls set_cached_projects (file write) + get_cached_projects (file read)
-        # each round. OS page-cache state on shared runners causes 3–5x variation between
+        # each round. OS page-cache state on shared runners causes 3-5x variation between
         # consecutive CI runs, making this ungatable with any reasonable slack.
         "test_summary_cache_round_trip",
         # Sub-100µs in-memory cache lookups vary 2.5x+ between consecutive ubuntu-latest
-        # runs; gated ratio band (0.5x–1.2x) cannot bracket both without false failures.
+        # runs; gated ratio band (0.5x to 1.2x) cannot bracket both without false failures.
         "test_summary_cache_lookup[hit]",
         "test_summary_cache_lookup[miss]",
         "test_composer_map_cache_lookup[hit]",

tests/test_check_benchmark_regression.py

@@ -7,6 +7,7 @@
 import pytest
 
 from scripts.check_benchmark_regression import (
+    EXCLUDED_FROM_GATE,
     BenchmarkDataError,
     check_regression,
     load_baseline_means,
@@ -16,6 +17,17 @@
 
 GATED_BENCH = "test_fingerprint_workspace_entries[10]"
 
+# Pytest benchmark node IDs (parametrize ids) that must stay in EXCLUDED_FROM_GATE.
+EXPECTED_EXCLUDED_FROM_GATE = (
+    "test_summary_cache_round_trip",
+    "test_summary_cache_lookup[hit]",
+    "test_summary_cache_lookup[miss]",
+    "test_composer_map_cache_lookup[hit]",
+    "test_composer_map_cache_lookup[miss]",
+    "test_tab_summary_cache_lookup[hit]",
+    "test_tab_summary_cache_lookup[miss]",
+)
+
 
 def _write_results(path, benchmarks: list[dict]) -> None:
     path.write_text(
@@ -31,6 +43,43 @@ def _write_baselines(path, groups: dict[str, dict[str, float]]) -> None:
     )
 
 
+def test_excluded_from_gate_matches_benchmark_ids() -> None:
+    assert EXCLUDED_FROM_GATE == frozenset(EXPECTED_EXCLUDED_FROM_GATE)
+
+
+@pytest.mark.parametrize("excluded_bench", EXPECTED_EXCLUDED_FROM_GATE)
+def test_excluded_benchmark_skips_regression_gate(
+    tmp_path, excluded_bench: str
+) -> None:
+    results = tmp_path / "results.json"
+    baselines = tmp_path / "baselines.json"
+    _write_results(
+        results,
+        [{"name": excluded_bench, "stats": {"mean": 1.0}}],
+    )
+    _write_baselines(
+        baselines,
+        {"summary-cache": {excluded_bench: 0.1}},
+    )
+
+    assert check_regression(results, baselines) == 0
+
+
+@pytest.mark.parametrize("excluded_bench", EXPECTED_EXCLUDED_FROM_GATE)
+def test_excluded_benchmark_missing_result_does_not_fail(
+    tmp_path, excluded_bench: str
+) -> None:
+    results = tmp_path / "results.json"
+    baselines = tmp_path / "baselines.json"
+    _write_results(results, [])
+    _write_baselines(
+        baselines,
+        {"summary-cache": {excluded_bench: 0.1}},
+    )
+
+    assert check_regression(results, baselines) == 0
+
+
 def test_normalize_benchmark_name_strips_module_prefix() -> None:
     full = "tests/benchmarks/test_summary_cache_bench.py::test_summary_cache_lookup[hit]"
     assert normalize_benchmark_name(full) == "test_summary_cache_lookup[hit]"

tests/test_models.py

@@ -93,6 +93,38 @@ def test_export_entry_parses(self) -> None:
         self.assertEqual(entry.workspace, "ws-1")
 
 
+class CursorStoragePayloadContract(unittest.TestCase):
+    """cursor_storage_payload() must shallow-copy _raw for API passthrough."""
+
+    def _assert_shallow_copy(self, model: Composer | Bubble | WorkspaceLocalComposer, raw: dict) -> None:
+        raw["nested"] = {"k": 1}
+        payload = model.cursor_storage_payload()
+        self.assertIsNot(payload, raw)
+        self.assertEqual(payload, raw)
+        self.assertIs(payload["nested"], raw["nested"])
+        payload["__api_mutation__"] = True
+        self.assertNotIn("__api_mutation__", raw)
+
+    def test_composer_cursor_storage_payload_is_shallow_copy(self) -> None:
+        raw = dict(GOOD_COMPOSER_RAW)
+        composer = Composer.from_dict(raw, composer_id="cid-001")
+        self._assert_shallow_copy(composer, raw)
+
+    def test_bubble_cursor_storage_payload_is_shallow_copy(self) -> None:
+        raw = {"text": "hi", "metadata": {"model": "gpt-4"}}
+        bubble = Bubble.from_dict(raw, bubble_id="b-1")
+        self._assert_shallow_copy(bubble, raw)
+
+    def test_workspace_local_composer_cursor_storage_payload_is_shallow_copy(self) -> None:
+        raw = {
+            "composerId": "cid-local-1",
+            "lastUpdatedAt": 1_715_000_500_000,
+            "conversation": [{"bubbleId": "b-1"}],
+        }
+        local = WorkspaceLocalComposer.from_dict(raw)
+        self._assert_shallow_copy(local, raw)
+
+
 class ComposerMissingFieldSchema(unittest.TestCase):
     def test_missing_full_conversation_headers_only_raises(self) -> None:
         bad = {k: v for k, v in GOOD_COMPOSER_RAW.items() if k != "fullConversationHeadersOnly"}