From 780758a6ef575179bd65552734229cbf7a6260dd Mon Sep 17 00:00:00 2001 From: Alpamys Date: Fri, 17 Jul 2026 16:31:43 +0500 Subject: [PATCH] test: record the MINI_MMLU coupling behind the 0.4 gate expectation (#315) The benchmark gate-task test pins score == 0.4, which is 2/5 -- exactly two MINI_MMLU answers are "B" and the fake generate_fn returns "B". Editing the fixture in forgetting.py moves the number, so say so. Comment only, no logic change. Co-Authored-By: Claude Opus 4.8 (1M context) --- tests/test_part_a_wave1.py | 2 ++ 1 file changed, 2 insertions(+) diff --git a/tests/test_part_a_wave1.py b/tests/test_part_a_wave1.py index 0d05548..d8fc79e 100644 --- a/tests/test_part_a_wave1.py +++ b/tests/test_part_a_wave1.py @@ -131,6 +131,8 @@ class TestRunGateErrorPropagation: result = run_gate(suite, generate_fn=lambda _p: "B") row = result.task_results[0] + # 0.4 == 2/5: exactly two MINI_MMLU answers are "B". Editing that + # fixture in forgetting.py moves this number. assert row.score == 0.4 assert row.error is None assert row.passed is True