fix(bench): harden reduce_baselines and fix Python 3.10 CI

Author: clean6378-max-it

benchmarks/baselines.json

@@ -1,5 +1,5 @@
 {
-  "_note": "Gated means from ubuntu-latest CI benchmark-results.json. Values multiplied by 1.5\u00d7 slack at generation time. Refresh after intentional speedups via reduce_baselines.py.",
+  "_note": "Gated means from ubuntu-latest CI benchmark-results.json. Values multiplied by 1.5x slack at generation time. Refresh after intentional speedups via reduce_baselines.py.",
   "updated": "2026-06-25T21:48:35Z",
   "machine": "Linux",
   "groups": {

scripts/reduce_baselines.py

@@ -6,7 +6,7 @@
 import json
 import math
 import sys
-from datetime import UTC, datetime
+from datetime import datetime, timezone
 from pathlib import Path
 
 _REPO_ROOT = Path(__file__).resolve().parent.parent
@@ -75,6 +75,11 @@ def reduce_baselines(
                 f"{path} benchmarks[{index}] ({bench_name!r}) has unknown group {group!r}; "
                 f"expected one of {GATED_GROUPS}"
             )
+        if bench_name in groups[group]:
+            raise BenchmarkDataError(
+                f"{path} benchmarks[{index}] ({raw_name!r}) duplicates normalized "
+                f"benchmark {group!r}/{bench_name!r}"
+            )
         groups[group][bench_name] = mean * slack
 
     excluded = ", ".join(sorted(EXCLUDED_FROM_GATE))
@@ -83,7 +88,7 @@ def reduce_baselines(
         if excluded
         else ""
     )
-    slack_note = f" Values multiplied by {slack}× slack at generation time." if slack != 1.0 else ""
+    slack_note = f" Values multiplied by {slack}x slack at generation time." if slack != 1.0 else ""
     machine_info = raw.get("machine_info")
     machine = machine_info.get("system") if isinstance(machine_info, dict) else None
     source_labels = {
@@ -97,7 +102,7 @@ def reduce_baselines(
             f"{slack_note}{excluded_note} "
             "Refresh after intentional speedups via reduce_baselines.py."
         ),
-        "updated": datetime.now(UTC).strftime("%Y-%m-%dT%H:%M:%SZ"),
+        "updated": datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ"),
         "machine": machine,
         "groups": groups,
     }

tests/test_reduce_baselines.py

@@ -59,9 +59,10 @@ def test_reduce_baselines_groups_and_slack(tmp_path) -> None:
     assert groups["parse"]["test_list_workspace_projects_nocache[composers-50]"] == pytest.approx(0.075)
     assert groups["export"]["test_post_export_zip[composers-10]"] == pytest.approx(0.015)
     assert groups["search"]["test_search_full_corpus"] == pytest.approx(0.06)
+    assert groups["summary-cache"]["test_summary_cache_lookup[hit]"] == pytest.approx(0.00015)
     assert data["machine"] == "Linux"
     assert "ubuntu-latest CI benchmark-results.json" in data["_note"]
-    assert "1.5× slack" in data["_note"]
+    assert "1.5x slack" in data["_note"]
     assert output["groups"] == groups
 
 
@@ -121,6 +122,29 @@ def test_reduce_baselines_rejects_missing_group(tmp_path) -> None:
         reduce_baselines(raw, out)
 
 
+def test_reduce_baselines_rejects_duplicate_normalized_name(tmp_path) -> None:
+    raw = tmp_path / "raw.json"
+    out = tmp_path / "baselines.json"
+    _write_raw(
+        raw,
+        [
+            {
+                "name": "test_summary_cache_lookup[hit]",
+                "group": "summary-cache",
+                "stats": {"mean": 0.0001},
+            },
+            {
+                "name": "tests/benchmarks/test_summary_cache_bench.py::test_summary_cache_lookup[hit]",
+                "group": "summary-cache",
+                "stats": {"mean": 0.0002},
+            },
+        ],
+    )
+
+    with pytest.raises(BenchmarkDataError, match="duplicates normalized"):
+        reduce_baselines(raw, out)
+
+
 def test_positive_float_rejects_non_finite() -> None:
     import argparse