diff --git a/CHANGELOG.md b/CHANGELOG.md
index a61be35..39ee24a 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -13,6 +13,13 @@ stability commitment. See [ROADMAP.md](ROADMAP.md) for planned milestones.
 
 ## [Unreleased]
 
+### Notes
+
+- `tests/test_iban_it.py::test_matching_time_on_long_runs` (a 50 ms budget
+  on 64k characters) carries the `benchmark` marker and is excluded from CI,
+  as the timing tests of `tests/test_firewall_pattern_timing.py` are. It
+  runs with `pytest -m benchmark tests/test_iban_it.py`.
+
 ## [0.13.1] — 2026-10-06
 
 Patch release: the compose file of `admina init` binds the dashboard, OTEL
diff --git a/tests/test_iban_it.py b/tests/test_iban_it.py
index 05c8919..8ac9563 100644
--- a/tests/test_iban_it.py
+++ b/tests/test_iban_it.py
@@ -208,6 +208,9 @@ def test_card_and_phone_numbers_in_one_text(regex_only):
 # ── Matching time on long inputs ──────────────────────────────
 
 
+# A time budget on 64k characters: run with `pytest -m benchmark` (CI runs
+# -m "not benchmark"; shared runners vary between runs).
+@pytest.mark.benchmark
 @pytest.mark.parametrize("unit", ["IT60", "IT60 ", "IT60X", "IT60 X", "A1 ", "1 ", "+39 3"])
 def test_matching_time_on_long_runs(regex_only, unit):
     from types import SimpleNamespace