diff --git a/backend/app/eval/score.py b/backend/app/eval/score.py
index f471c750..76e78307 100644
--- a/backend/app/eval/score.py
+++ b/backend/app/eval/score.py
@@ -21,7 +21,8 @@ wrong 错、scale 量纲不同（差 10 的整数次幂，比如元和万元、
 - table：逐行按 keys 的取值配对，每个要比的值同 scalar；ordered 时按答案里各行键值出现的先后核对顺序。
 - set：名单里每一项都要在答案里出现（ordered 时按顺序）；每一项都出现在答案查过的结果里算可追溯。
 - none：期望答「没有数据」：答案里没有数，而且查过的查询里有返回 0 行的（可追溯）；没有数、但也没有一条查询返回 0 行的
-  算「对但没引用」。
+  算「对但没引用」。「给了数」只认引用的数值格和叙述里写的数：引用文字格（地区名这种维度）、空值格（NULL）不算——「华中区
+  没有数据」引用了地区那一格，说的仍是没有数据。
 - ask：期望反问：作答路径是反问才对。
 - text：文字要点，由裁判判（covered / partial / missing / unjudged），单列「模型判断」，不进准确率（评测-D4）。
 
@@ -46,8 +47,10 @@ from app.engine.tolerance import agree, is_finite_number
 
 logger = logging.getLogger(__name__)
 
-#: 评分规则的版本：规则改了就加一，结果里记着按哪一版评的，可以按新规则重评
-SCORER_VERSION = 1
+#: 评分规则的版本：规则改了就加一，结果里记着按哪一版评的，可以按新规则重评。
+#: 2：期望「没有数据」时引用文字格、空值格不算给了数；判错时「答案里的数」给配对上的那一格；答案没引用查询时逐题 E1
+#: 不硬比第一条查询
+SCORER_VERSION = 2
 
 OUTCOMES = ("traced", "untraced", "wrong", "scale", "ungradable")
 OUTCOME_LABEL = {"traced": "对且可追溯", "untraced": "对但没引用", "wrong": "错", "scale": "量纲不同",
@@ -371,8 +374,15 @@ def _score_set(exp: Mapping[str, Any], gold: Mapping[str, Any], answer: Answer,
 
 
 def _score_none(answer: Answer) -> dict[str, Any]:
-    if answer.values or answer.bare:
-        return {"outcome": "wrong", "reason": "期望答「没有数据」，答案里却给了数"}
+    """期望答「没有数据」。「给了数」只认引用的数值格、叙述里写的数：引用的文字格（地区名这种维度）、空值格（NULL）不是数——
+    「华中区（[[v:Q1.r0.地区]]）没有数据」说的正是没有数据，以前因为它引用了一格就判成「给了数」。"""
+    numeric = next((v for v in answer.values if is_finite_number(v.get("value"))), None)
+    if numeric is not None:
+        return {"outcome": "wrong", "got": numeric.get("value"), "cell": _cell_ref(numeric),
+                "reason": "期望答「没有数据」，答案里却引用了数"}
+    if answer.bare:
+        return {"outcome": "wrong", "got": answer.bare[0]["value"], "raw": answer.bare[0]["raw"],
+                "reason": "期望答「没有数据」，答案的叙述里却写了数"}
     if any(int(q.get("row_count") or 0) == 0 and q.get("role") != "verify" for q in answer.queries):
         return {"outcome": "traced"}
     return {"outcome": "untraced", "reason": "答案没有给数，但没有一条查询返回 0 行可以说明「没有数据」"}
diff --git a/backend/tests/test_eval_score.py b/backend/tests/test_eval_score.py
index b9578a20..a4946a2d 100644
--- a/backend/tests/test_eval_score.py
+++ b/backend/tests/test_eval_score.py
@@ -116,6 +116,27 @@ def test_tables_sets_none_and_ask():
     assert score_case(ask, {"status": "ok"}, Answer(run_id="r"), path="ran").verdict == "fail"
 
 
+def test_none_counts_only_numbers_as_giving_data():
+    """期望「没有数据」：只有引用了数值格、或叙述里写了数，才算答案给了数。引用的是文字格（地区名）、空值格（NULL）的，
+    答案明说没有数据就按没有数据判对。"""
+    none = _case({"id": "n", "kind": "none", "required": True})
+    gold = _gold({"n": {"kind": "none", "rows": 0}})
+    empty = [{"artifact": "q", "row_count": 0, "sql": "SELECT 1"}]
+    text_cell = {"value": "华中", "kind": "cell", "artifact": "q", "row": 0, "column": "region", "dims": {}}
+    named = score_case(none, gold, Answer(run_id="r", text="华中区没有数据", values=[text_cell], queries=empty), path="ran")
+    assert named.verdict == "pass" and named.scores[0]["outcome"] == "traced", named.scores
+    null_cell = {"value": None, "kind": "cell", "artifact": "q", "row": 0, "column": "gmv", "dims": {}}
+    null = score_case(none, gold, Answer(run_id="r", text="华中区的销售额为空，没有数据", values=[null_cell],
+                                         queries=empty), path="ran")
+    assert null.verdict == "pass", null.scores
+    number = score_case(none, gold, Answer(run_id="r", text="华中区 12 元", values=[_cell(12.0)], queries=empty),
+                        path="ran")
+    assert number.verdict == "fail" and number.scores[0]["got"] == 12.0, number.scores
+    said = score_case(none, gold, Answer(run_id="r", text="华中区 12 元", values=[text_cell], bare=_bare("12 元"),
+                                         queries=empty), path="ran")
+    assert said.verdict == "fail" and said.scores[0]["raw"] == "12", said.scores
+
+
 def test_the_path_and_the_gold_decide_before_numbers():
     case = _case(EXP)
     gold = _gold({"e1": _scalar(330.0)})