全量修复:预测系统正确性、安全性与部署问题

P0 严重问题修复: - 修复 form_slice/stats_slice 主客身份反转(历史比赛视角错误) - 修复 understat.py httpx 未导入导致的 NameError - 修复 LLM 解析失败时静默产生假成功预测(0-0 平局+置信度0.5) 预测路径修复: - multi-agent 路径增加 backtest cutoff 透传,回测防泄漏生效 - H2H 切片汇总统计改为从当前主队视角计数 - 预测唯一约束增加 mode+run_type 维度,防止回测覆盖实盘预测伤停管线修复: - IntegrityError 后不再整批回滚丢数据(改用逐条 flush) - return_date 正确解析并写入 - retrieved_at 比较统一用 date() 避免当天数据不可见 - 唯一索引改为 partial unique index(排除 NULL 重复) - HTTP 缓存 TTL 从 7 天改为 6 小时安全与连接管理: - /api/v1/predict 增加内存滑动窗口限流(10次/分钟/IP) - 预测路由改用短 session 模式,LLM 调用期间不持有 DB 连接 Docker 部署修复: - 修复 .dockerignore 排除 *.md 导致 COPY README.md 失败 - 容器内 DATABASE_URL 使用 postgres 服务名(非 localhost) - 启动时自动执行 alembic upgrade head - 前端改用多阶段构建(Dockerfile.frontend) 新增测试(5个文件,24+用例): - test_p0_home_away.py: 主客身份反转回归测试 - test_p0_parse_failure.py: LLM 解析失败回归测试 - test_multi_agent_cutoff.py: multi-agent cutoff 透传测试 - test_h2h_perspective.py: H2H 视角测试 - test_injuries_pipeline.py: 伤停管线 5 项修复测试 - test_predict_protection.py: 限流+短 session 测试 - test_prediction_unique_constraint.py: 唯一约束测试迁移: - 0012_injuries_partial_unique_and_return_date.py - 0013_predictions_unique_constraint_mode_run_type.py
2026-09-19 06:43:55 +00:00
parent 11efe91ce9
commit bee330f31f
27 changed files with 1666 additions and 137 deletions
@@ -0,0 +1,114 @@
+"""回归测试: P0-3 — LLM 解析失败不能产生假成功预测。
+
+验证链路:
+  1. provider.py: JSON 解析失败时必须设置 error
+  2. predict.py: resp.parsed is None 时必须抛错,不能 fallback 到 {}
+  3. validation.py: 必填字段缺失时必须失败,不能静默给默认值
+"""
+from __future__ import annotations
+
+from unittest.mock import MagicMock
+
+import pytest
+
+from src.llm.provider import LLMProvider, LLMResponse
+from src.llm.validation import validate_prediction_output
+
+
+class TestProviderJsonParseError:
+    """P0-3 Part 1: provider.py JSON 解析失败必须设置 error。"""
+
+    @pytest.mark.asyncio
+    async def test_invalid_json_sets_error(self, monkeypatch):
+        """LLM 返回非 JSON 内容时,error 必须非空。"""
+        async def fake_post(*args, **kwargs):
+            class FakeResp:
+                status_code = 200
+                def raise_for_status(self): pass
+                def json(self):
+                    return {
+                        "choices": [{"message": {"content": "我不确定,可能是平局"}}],
+                        "usage": {"prompt_tokens": 10, "completion_tokens": 5},
+                    }
+            return FakeResp()
+
+        import httpx
+        monkeypatch.setattr(httpx.AsyncClient, "post", fake_post)
+
+        p = LLMProvider(api_key="test", model="gpt-4o")
+        resp = await p.chat("sys", "user", json_mode=True)
+        # P0-3: JSON 解析失败必须设置 error
+        assert resp.error is not None, "JSON 解析失败应设置 error"
+        assert resp.parsed is None
+
+    @pytest.mark.asyncio
+    async def test_code_block_json_works(self, monkeypatch):
+        """LLM 返回 ```json {...}}``` 时应成功解析。"""
+        async def fake_post(*args, **kwargs):
+            class FakeResp:
+                status_code = 200
+                def raise_for_status(self): pass
+                def json(self):
+                    return {
+                        "choices": [{"message": {"content": '```json\n{"pred_home_goals": 1.5, "pred_away_goals": 1.0, "pred_1x2": "1", "subjective_confidence": 0.7}\n```'}}],
+                        "usage": {"prompt_tokens": 10, "completion_tokens": 5},
+                    }
+            return FakeResp()
+
+        import httpx
+        monkeypatch.setattr(httpx.AsyncClient, "post", fake_post)
+
+        p = LLMProvider(api_key="test", model="gpt-4o")
+        resp = await p.chat("sys", "user", json_mode=True)
+        assert resp.error is None
+        assert resp.parsed is not None
+        assert resp.parsed["pred_1x2"] == "1"
+
+
+class TestValidationNoSilentDefaults:
+    """P0-3 Part 3: validation.py 必填字段缺失时必须失败。"""
+
+    def test_missing_pred_home_goals_raises(self):
+        """缺少 pred_home_goals 必须报错,不能默认为 0。"""
+        with pytest.raises((ValueError, KeyError)):
+            validate_prediction_output({
+                "pred_away_goals": 1,
+                "pred_1x2": "1",
+                "subjective_confidence": 0.7,
+            })
+
+    def test_missing_pred_1x2_raises(self):
+        """缺少 pred_1x2 必须报错,不能默认为 X。"""
+        with pytest.raises(ValueError, match="Missing required field: pred_1x2"):
+            validate_prediction_output({
+                "pred_home_goals": 1,
+                "pred_away_goals": 0,
+                "subjective_confidence": 0.7,
+            })
+
+    def test_missing_confidence_raises(self):
+        """缺少 subjective_confidence 必须报错,不能默认为 0.5。"""
+        with pytest.raises(ValueError, match="Missing required field: subjective_confidence"):
+            validate_prediction_output({
+                "pred_home_goals": 1,
+                "pred_away_goals": 0,
+                "pred_1x2": "1",
+            })
+
+    def test_empty_dict_raises(self):
+        """空 dict 必须报错(不能产生 0-0 X 0.5 的假预测)。"""
+        with pytest.raises((ValueError, KeyError)):
+            validate_prediction_output({})
+
+    def test_valid_input_passes(self):
+        """完整的合法输入应通过。"""
+        result = validate_prediction_output({
+            "pred_home_goals": 1.5,
+            "pred_away_goals": 1.0,
+            "pred_1x2": "1",
+            "subjective_confidence": 0.7,
+        })
+        assert result.pred_home_goals == 2  # 1.5 → round → 2
+        assert result.pred_away_goals == 1
+        assert result.pred_1x2 == "1"
+        assert result.subjective_confidence == 0.7