完善评估能力：筛选参数 + degraded 排除 + 前端评估页

后端：
- settle_prediction 拒绝 degraded/failed（明确错误信息）
- get_eval_summary 支持 provider/model/prompt_version/mode 筛选
- 返回 filtered_settled/evaluated/skipped_degraded 等计数
- matches 游标分页方向修复（scheduled ASC 用 > 条件）
- available_at 加 2h 缓冲（近似完赛时间）
- bzzziro 统计字段映射注释（待真实响应验证）
- injuries 区分 no_local_data 与 success 空名单

前端：
- 新增 EvalPage（筛选控件 + 汇总卡片 + 准确率表格）
- 挂载 /admin/eval 路由与导航

测试：
- test_matches_cursor.py：游标方向
- test_available_at.py：2h 缓冲与回测防泄漏
- test_bzzoirot_stats.py：统计字段映射
- test_injuries_no_local_data.py：no_local_data vs success
- test_injuries_inserted_count.py：失败批不计入
- test_eval_excludes_degraded.py：degraded 排除准确率

tests/test_matches_cursor.py

@@ -0,0 +1,89 @@
+"""回归测试: 比赛列表游标分页方向修复。
+
+验证:
+  - status=scheduled 时,游标条件为「大于」(ASC 方向)
+  - 其它 status 时,游标条件为「小于」(DESC 方向)
+  - 无 cursor 时行为不变
+"""
+from __future__ import annotations
+
+from datetime import datetime, timezone
+
+import pytest
+from sqlalchemy import select
+
+from src.db.models import Match
+
+
+class TestCursorPaginationDirection:
+    """验证游标条件方向与排序方向一致。"""
+
+    def _build_query(self, status=None, cursor=None):
+        """复现 list_matches 的查询构造逻辑,返回 where 条件列表。"""
+        q = select(Match)
+
+        if cursor:
+            last_date_str, last_id_str = cursor.split("|", 1)
+            last_date = datetime.fromisoformat(last_date_str)
+            last_id = int(last_id_str)
+            if status == "scheduled":
+                q = q.where(
+                    (Match.match_date > last_date) |
+                    ((Match.match_date == last_date) & (Match.id > last_id))
+                )
+            else:
+                q = q.where(
+                    (Match.match_date < last_date) |
+                    ((Match.match_date == last_date) & (Match.id < last_id))
+                )
+
+        if status:
+            q = q.where(Match.match_status == status)
+
+        if status == "scheduled":
+            order = (Match.match_date.asc(), Match.id.asc())
+        else:
+            order = (Match.match_date.desc(), Match.id.desc())
+
+        return q.order_by(*order)
+
+    def test_scheduled_uses_greater_than(self):
+        """scheduled + cursor: 应使用 > 条件(ASC 方向)。"""
+        q = self._build_query(
+            status="scheduled",
+            cursor="2026-01-15T15:00:00|100"
+        )
+        sql = str(q)
+        assert ">" in sql, f"scheduled 游标应使用 >,SQL: {sql}"
+        assert "<" not in sql or "match_date <" not in sql, f"不应出现 < 条件"
+
+    def test_other_status_uses_less_than(self):
+        """finished + cursor: 应使用 < 条件(DESC 方向)。"""
+        q = self._build_query(
+            status="finished",
+            cursor="2026-01-15T15:00:00|100"
+        )
+        sql = str(q)
+        assert "<" in sql, f"finished 游标应使用 <,SQL: {sql}"
+        assert "match_date >" not in sql, f"不应出现 > 条件"
+
+    def test_no_cursor_no_direction(self):
+        """无 cursor 时不应有游标条件。"""
+        q = self._build_query(status="scheduled", cursor=None)
+        sql = str(q)
+        # 应无 match_date 比较条件(只有 status filter)
+        assert "match_date >" not in sql
+        assert "match_date <" not in sql
+
+    def test_scheduled_order_is_asc(self):
+        """scheduled 排序应为 ASC。"""
+        q = self._build_query(status="scheduled", cursor=None)
+        sql = str(q)
+        assert "ASC" in sql, f"scheduled 应 ASC 排序,SQL: {sql}"
+        assert "DESC" not in sql, f"不应出现 DESC"
+
+    def test_finished_order_is_desc(self):
+        """finished 排序应为 DESC。"""
+        q = self._build_query(status="finished", cursor=None)
+        sql = str(q)
+        assert "DESC" in sql, f"finished 应 DESC 排序,SQL: {sql}"