feat(harness): add task_types, pool_split_mode, train_ratio, test_questions to RunConfig

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

tests/unit/test_harness_config.py

@@ -600,3 +600,76 @@ def test_video_mme_task_type_count_is_12():
     from app.harness.config import _VIDEO_MME_TASK_TYPE_COUNT
 
     assert _VIDEO_MME_TASK_TYPE_COUNT == 12
+
+
+# ────────────────────── test_run_config_new_fields ─────────────────────────
+
+
+class TestRunConfigNewFields:
+    """RunConfig 新增字段校验（task_types / pool_split_mode / train_ratio / test_questions）。"""
+
+    def test_pool_split_mode_valid(self) -> None:
+        """pool_split_mode 合法值应通过校验。"""
+        for mode in ("global", "per_category"):
+            cfg = _make_config(pool_split_mode=mode)
+            _validate(cfg)
+
+    def test_pool_split_mode_invalid(self) -> None:
+        """pool_split_mode 非法值应报错。"""
+        cfg = _make_config(pool_split_mode="invalid")
+        with pytest.raises(ValueError, match="pool_split_mode"):
+            _validate(cfg)
+
+    def test_task_types_tuple(self) -> None:
+        """task_types 接受 tuple。"""
+        cfg = _make_config(task_types=("short", "medium"))
+        _validate(cfg)
+        assert cfg.task_types == ("short", "medium")
+
+    def test_task_types_none(self) -> None:
+        """task_types 默认 None。"""
+        cfg = _make_config()
+        assert cfg.task_types is None
+
+    def test_train_ratio_valid(self) -> None:
+        """train_ratio 在 (0, 1) 内应通过。"""
+        cfg = _make_config(train_ratio=0.5)
+        _validate(cfg)
+
+    def test_train_ratio_zero_rejected(self) -> None:
+        """train_ratio <= 0 应报错。"""
+        cfg = _make_config(train_ratio=0.0)
+        with pytest.raises(ValueError, match="train_ratio"):
+            _validate(cfg)
+
+    def test_train_ratio_one_rejected(self) -> None:
+        """train_ratio >= 1 应报错。"""
+        cfg = _make_config(train_ratio=1.0)
+        with pytest.raises(ValueError, match="train_ratio"):
+            _validate(cfg)
+
+    def test_train_ratio_negative_rejected(self) -> None:
+        """train_ratio < 0 应报错。"""
+        cfg = _make_config(train_ratio=-0.1)
+        with pytest.raises(ValueError, match="train_ratio"):
+            _validate(cfg)
+
+    def test_test_questions_default(self) -> None:
+        """test_questions 默认值为 benchmarks/Video-MME。"""
+        cfg = _make_config()
+        assert cfg.test_questions == "benchmarks/Video-MME"
+
+    def test_per_category_skips_val_size_check(self) -> None:
+        """per_category 模式跳过 val_size >= eval_min_per_class * 12 的校验。
+
+        val_size=1 在 global 模式会因低于 floor 而报错，
+        但 per_category 模式应跳过该检查。
+        """
+        cfg = _make_config(pool_split_mode="per_category", val_size=1)
+        _validate(cfg)
+
+    def test_global_mode_enforces_val_size_check(self) -> None:
+        """global 模式仍然执行 val_size floor 检查。"""
+        cfg = _make_config(pool_split_mode="global", eval_min_per_class=3, val_size=30)
+        with pytest.raises(ValueError, match="val_size"):
+            _validate(cfg)