diff --git a/src/api/routes/predict.py b/src/api/routes/predict.py
index bcf99c1..eff6e84 100644
--- a/src/api/routes/predict.py
+++ b/src/api/routes/predict.py
@@ -63,7 +63,8 @@ async def predict(req: PredictRequest, request: Request):
         logger.exception("predict unexpected error")
         raise HTTPException(500, "预测失败,请查看服务器日志")
 
-    # baseline 模式:结果已是 dict,需独立落库(prediction_id)
+    # D2: 三种模式统一返回 PredictResult —— 字段映射单一化,无 dict 分支。
+    # 仅 baseline 的 prediction_id 需要在此落库补齐(服务层不落库)。
     if req.mode == "baseline":
         prediction_id = await _persist_baseline(req.match_id, result)
     else:
@@ -73,38 +74,34 @@ async def predict(req: PredictRequest, request: Request):
     logger.info(
         "预测完成 match=%s mode=%s pred=%s:%s (%s)",
         req.match_id, req.mode,
-        result.get("pred_home_goals") if isinstance(result, dict) else result.pred_home_goals,
-        result.get("pred_away_goals") if isinstance(result, dict) else result.pred_away_goals,
-        result.get("pred_1x2") if isinstance(result, dict) else result.pred_1x2,
+        result.pred_home_goals, result.pred_away_goals, result.pred_1x2,
     )
 
-    result_dict = result if isinstance(result, dict) else None
-
     return PredictOut(
         prediction_id=prediction_id,
-        provider=result.get("provider") if result_dict else result.provider,
-        model=result.get("model") if result_dict else result.model,
-        prompt_version=result.get("prompt_version") if result_dict else getattr(result, "prompt_version", None),
+        provider=result.provider,
+        model=result.model,
+        prompt_version=result.prompt_version,
         mode=req.mode,
-        pred_home_goals=result.get("pred_home_goals") if result_dict else result.pred_home_goals,
-        pred_away_goals=result.get("pred_away_goals") if result_dict else result.pred_away_goals,
-        alt_pred_home_goals=result.get("alt_pred_home_goals") if result_dict else result.alt_pred_home_goals,
-        alt_pred_away_goals=result.get("alt_pred_away_goals") if result_dict else result.alt_pred_away_goals,
-        pred_1x2=result.get("pred_1x2") if result_dict else result.pred_1x2,
-        subjective_confidence=result.get("subjective_confidence") if result_dict else result.subjective_confidence,
-        reasoning=result.get("reasoning") if result_dict else result.reasoning,
-        status=result.get("status", "success") if result_dict else getattr(result, "status", "success"),
-        agent_outputs=result.get("agent_outputs") if result_dict else getattr(result, "agent_outputs", None),
-        agent_weights=result.get("agent_weights") if result_dict else getattr(result, "agent_weights", None),
-        context=result.get("context", "") if result_dict else result.context,
-        latency_ms=result.get("latency_ms", 0) if result_dict else result.latency_ms,
-        prompt_tokens=result.get("prompt_tokens") if result_dict else getattr(result, "prompt_tokens", None),
-        completion_tokens=result.get("completion_tokens") if result_dict else getattr(result, "completion_tokens", None),
+        pred_home_goals=result.pred_home_goals,
+        pred_away_goals=result.pred_away_goals,
+        alt_pred_home_goals=result.alt_pred_home_goals,
+        alt_pred_away_goals=result.alt_pred_away_goals,
+        pred_1x2=result.pred_1x2,
+        subjective_confidence=result.subjective_confidence,
+        reasoning=result.reasoning,
+        status=result.status,
+        agent_outputs=result.agent_outputs,
+        agent_weights=result.agent_weights,
+        context=result.context,
+        latency_ms=result.latency_ms,
+        prompt_tokens=result.prompt_tokens,
+        completion_tokens=result.completion_tokens,
         rate_limit_remaining=get_predict_rate_limit_remaining(request),
     )
 
 
-async def _persist_baseline(match_id: int, baseline: dict) -> int:
+async def _persist_baseline(match_id: int, baseline: PredictResult) -> int:
     """将基线预测结果写入 prediction 表,复用 upsert 语义。"""
     from src.db.unit_of_work import get_uow
     from src.llm.predict import _upsert_prediction
@@ -118,16 +115,16 @@ async def _persist_baseline(match_id: int, baseline: dict) -> int:
             mode="baseline",
             run_type="live",  # baseline 是 live 预测的变体,符合 ck_run_type_enum
             values={
-                "prompt_version": "baseline_v1",
-                "prompt_tokens": 0,
-                "completion_tokens": 0,
-                "latency_ms": 0,
-                "pred_home_goals": baseline["pred_home_goals"],
-                "pred_away_goals": baseline["pred_away_goals"],
-                "pred_1x2": baseline["pred_1x2"],
-                "subjective_confidence": baseline["subjective_confidence"],
-                "reasoning": baseline["reasoning"],
-                "raw_response": baseline.get("raw", baseline),
+                "prompt_version": baseline.prompt_version,
+                "prompt_tokens": baseline.prompt_tokens or 0,
+                "completion_tokens": baseline.completion_tokens or 0,
+                "latency_ms": baseline.latency_ms or 0,
+                "pred_home_goals": baseline.pred_home_goals,
+                "pred_away_goals": baseline.pred_away_goals,
+                "pred_1x2": baseline.pred_1x2,
+                "subjective_confidence": baseline.subjective_confidence,
+                "reasoning": baseline.reasoning,
+                "raw_response": baseline.raw,
                 "status": "success",
             },
         )
diff --git a/src/llm/agents/orchestrator.py b/src/llm/agents/orchestrator.py
index f54621b..d1b214d 100644
--- a/src/llm/agents/orchestrator.py
+++ b/src/llm/agents/orchestrator.py
@@ -6,14 +6,13 @@ import hashlib
 import json
 import logging
 import time
-from dataclasses import dataclass
 from datetime import datetime, timezone
 
 from src.core.config import settings
 from src.db.base import AsyncSessionLocal
 from src.db.models import Match, Prediction
 from src.db.unit_of_work import get_uow
-from src.llm.predict import _upsert_prediction
+from src.llm.predict import PredictResult, _upsert_prediction
 from src.llm.agents.base import AgentReport, AgentSpec, load_agent_prompt
 from src.llm.context_builder import (
     MatchHeader,
@@ -81,28 +80,12 @@ AGENT_LABELS_ZH: dict[str, str] = {
 }
 
 
-@dataclass
-class MultiPredictResult:
-    prediction_id: int
-    provider: str
-    model: str
-    prompt_version: str
-    mode: str
-    pred_home_goals: float | None
-    pred_away_goals: float | None
-    alt_pred_home_goals: int | None
-    alt_pred_away_goals: int | None
-    pred_1x2: str | None
-    subjective_confidence: float | None
-    reasoning: str | None
-    context: str
-    agent_outputs: list[dict]
-    agent_weights: dict | None
-    status: str = "success"
-    latency_ms: int | None = None
-    prompt_tokens: int | None = None
-    completion_tokens: int | None = None
-    raw: dict | None = None
+# D2(工程债): multi 结果类型与 single 统一 —— 扩展后的 PredictResult 用可选
+# 字段(agent_outputs/agent_weights/prompt_tokens/completion_tokens/mode)承载
+# 全部模式,此处仅保留别名。保留 `MultiPredictResult` 名字的原因:
+#   1. predict_match_multi 签名 `-> MultiPredictResult:` 是 R4 源码守卫的标记;
+#   2. src/llm/agents/__init__.py 对外 re-export 该名字。
+MultiPredictResult = PredictResult
 
 
 async def _agent_provider(agent_id: str, *, tier: str, model_override: str | None = None) -> LLMProvider:
diff --git a/src/llm/baseline.py b/src/llm/baseline.py
index baad4c7..d4aa690 100644
--- a/src/llm/baseline.py
+++ b/src/llm/baseline.py
@@ -12,6 +12,7 @@ from sqlalchemy import case, func, select
 
 from src.db.base import AsyncSession, AsyncSessionLocal
 from src.db.models import Match
+from src.llm.predict import PredictResult
 
 logger = logging.getLogger(__name__)
 
@@ -52,11 +53,13 @@ async def predict_baseline(
     *,
     backtest: bool = False,
     cutoff_at: datetime | None = None,
-) -> dict:
+) -> PredictResult:
     """极简基线预测:主场场均进球 vs 客场场均进球。
 
-    返回与 PredictResult 兼容的字典:
+    返回 PredictResult(D2 统一结果类型):
       provider=model="baseline", 不调用 LLM,latency_ms≈0。
+    prediction_id 为占位 0 —— baseline 不在服务层落库,
+    由路由层 _persist_baseline 落库后取得真实 id。
     """
     async with AsyncSessionLocal() as db:
         match = await db.get(Match, match_id)
@@ -90,24 +93,26 @@ async def predict_baseline(
     else:
         pred_1x2 = "X"
 
-    return {
-        "pred_home_goals": float(pred_home),
-        "pred_away_goals": float(pred_away),
-        "alt_pred_home_goals": None,
-        "alt_pred_away_goals": None,
-        "pred_1x2": pred_1x2,
-        "subjective_confidence": 0.5,
-        "prompt_tokens": 0,
-        "completion_tokens": 0,
-        "reasoning": (
+    return PredictResult(
+        prediction_id=0,  # 占位:真实 id 由路由层 _persist_baseline 落库后返回
+        provider="baseline",
+        model="baseline",
+        prompt_version="baseline_v1",
+        mode="baseline",
+        pred_home_goals=float(pred_home),
+        pred_away_goals=float(pred_away),
+        alt_pred_home_goals=None,
+        alt_pred_away_goals=None,
+        pred_1x2=pred_1x2,
+        subjective_confidence=0.5,
+        reasoning=(
             f"基线估计(非投注建议): 主队主场场均进球 {home_avg:.2f} → 预测 {pred_home}; "
             f"客队客场场均进球 {away_avg:.2f} → 预测 {pred_away}。"
         ),
-        "provider": "baseline",
-        "model": "baseline",
-        "prompt_version": "baseline_v1",
-        "mode": "baseline",
-        "status": "success",
-        "latency_ms": 0,
-        "raw": {"home_avg": round(home_avg, 2), "away_avg": round(away_avg, 2)},
-    }
+        context="",  # baseline 不构建 LLM 上下文
+        status="success",
+        latency_ms=0,
+        prompt_tokens=0,
+        completion_tokens=0,
+        raw={"home_avg": round(home_avg, 2), "away_avg": round(away_avg, 2)},
+    )
diff --git a/src/llm/predict.py b/src/llm/predict.py
index 383b79f..c302e95 100644
--- a/src/llm/predict.py
+++ b/src/llm/predict.py
@@ -89,6 +89,14 @@ def _prompt_template_hash(version: str) -> str:
 
 @dataclass
 class PredictResult:
+    """三种预测模式(single/multi/baseline)的统一结果类型。
+
+    D2(工程债): 原本 single 返回本类、multi 重复定义 MultiPredictResult、
+    baseline 返回裸 dict,导致路由 isinstance(dict) 双分支 + backtest 对
+    baseline 直接 AttributeError。现以可选字段扩展本类承载全部模式;
+    MultiPredictResult 是本类的别名(见 src/llm/agents/orchestrator.py)。
+    """
+
     prediction_id: int
     provider: str
     model: str
@@ -101,8 +109,15 @@ class PredictResult:
     subjective_confidence: float | None
     reasoning: str | None
     context: str
+    # 模式标识: single(默认) / multi / baseline
+    mode: str = "single"
+    # multi 专属: 各专家报告列表与融合权重;single/baseline 为 None
+    agent_outputs: list[dict] | None = None
+    agent_weights: dict | None = None
     status: str = "success"
     latency_ms: int | None = None
+    prompt_tokens: int | None = None
+    completion_tokens: int | None = None
     raw: dict | None = None
 
 
@@ -158,9 +173,11 @@ async def predict_match(
     use_cache: bool = True,
     backtest: bool = False,
     cutoff_at=None,
-) -> "PredictResult | MultiPredictResult":
+) -> PredictResult:
     """预测入口。mode=multi(默认)走多 agent;mode=single 走单次调用;mode=baseline 走无 LLM 基线。
 
+    三种模式统一返回 PredictResult(D2);multi 的 MultiPredictResult 是其别名。
+
     Args:
         mode: multi(默认,5 专家+终裁) / single(单次) / baseline(极简统计基线,不调用 LLM)。
         use_cache:是否允许返回进程内缓存结果。回测必须传 False——
diff --git a/tests/test_baseline.py b/tests/test_baseline.py
index 6d76338..2eb7564 100644
--- a/tests/test_baseline.py
+++ b/tests/test_baseline.py
@@ -81,18 +81,18 @@ async def test_predict_baseline_no_llm():
 
         result = await predict_baseline(1)
 
-    assert result["provider"] == "baseline"
-    assert result["model"] == "baseline"
-    assert result["mode"] == "baseline"
-    assert result["latency_ms"] == 0
-    assert result["prompt_tokens"] == 0
-    assert result["completion_tokens"] == 0
+    assert result.provider == "baseline"
+    assert result.model == "baseline"
+    assert result.mode == "baseline"
+    assert result.latency_ms == 0
+    assert result.prompt_tokens == 0
+    assert result.completion_tokens == 0
     # 2.4 → round = 2, 1.6 → round = 2 → 平局 X
-    assert result["pred_home_goals"] == 2.0
-    assert result["pred_away_goals"] == 2.0
-    assert result["pred_1x2"] == "X"
-    assert result["subjective_confidence"] == 0.5
-    assert "非投注建议" in result["reasoning"]
+    assert result.pred_home_goals == 2.0
+    assert result.pred_away_goals == 2.0
+    assert result.pred_1x2 == "X"
+    assert result.subjective_confidence == 0.5
+    assert "非投注建议" in result.reasoning
     # 确认未调用任何 LLM 相关模块
     assert "home_10" in captured and "away_20" in captured
 
@@ -125,6 +125,6 @@ async def test_predict_baseline_clamps_to_range():
 
         result = await predict_baseline(2)
 
-    assert result["pred_home_goals"] == 10.0  # clamped
-    assert result["pred_away_goals"] == 0.0   # clamped
-    assert result["pred_1x2"] == "1"         # 10:0 主胜
+    assert result.pred_home_goals == 10.0  # clamped
+    assert result.pred_away_goals == 0.0   # clamped
+    assert result.pred_1x2 == "1"         # 10:0 主胜
diff --git a/tests/test_d2_predict_unification.py b/tests/test_d2_predict_unification.py
new file mode 100644
index 0000000..aa9a386
--- /dev/null
+++ b/tests/test_d2_predict_unification.py
@@ -0,0 +1,216 @@
+"""D2 工程债回归测试: 统一预测结果类型。
+
+背景: predict_match 三条路径返回类型不一 —— single 返回 PredictResult,
+multi 返回字段重复定义的 MultiPredictResult dataclass,baseline 返回裸 dict。
+后果: (1) 预测路由 PredictOut 映射被迫写 isinstance(result, dict) 双分支;
+(2) backtest 对 baseline 模式直接 AttributeError(dict 没有 .prediction_id,
+潜伏 bug);(3) 字段清单在两处 dataclass 重复维护,加字段必漏一处。
+
+统一方案: 扩展 PredictResult(可选字段)承载全部模式;
+MultiPredictResult 变为其别名(保留 orchestrator 签名标记,兼容 re-export);
+baseline 返回 PredictResult;路由单一字段映射。
+
+本测试守护四件事:
+  1. predict_baseline 返回 PredictResult(属性访问)
+  2. MultiPredictResult 与 PredictResult 兼容(orchestrator 构造调用的
+     全字段 kwargs 可直接构造别名)
+  3. 预测路由不再有 isinstance(result, dict) 分支(源码守卫,仿 R4 范式)
+  4. _persist_baseline 用属性访问构造 upsert values(baseline 落库语义不变)
+"""
+from __future__ import annotations
+
+from pathlib import Path
+from types import SimpleNamespace
+
+import pytest
+
+from src.llm.baseline import predict_baseline
+from src.llm.predict import PredictResult
+from src.llm.agents import MultiPredictResult
+
+ROUTE_PATH = Path(__file__).resolve().parents[1] / "src" / "api" / "routes" / "predict.py"
+
+
+# ============================================================
+# 1. baseline 返回 PredictResult
+# ============================================================
+
+
+@pytest.mark.asyncio
+async def test_predict_baseline_returns_predict_result():
+    """基线预测返回 PredictResult 实例,mode=baseline,token/延迟为 0。"""
+    from unittest.mock import patch
+
+    async def fake_avg(db, *, team_id, side, league_id, before):
+        return 2.4 if side == "home" else 1.6
+
+    class FakeMatch:
+        id = 1
+        home_team_id = 10
+        away_team_id = 20
+        league_id = 1
+        match_status = "scheduled"
+
+    class FakeSession:
+        async def get(self, cls, mid):
+            return FakeMatch()
+
+    class FakeCM:
+        async def __aenter__(self):
+            return FakeSession()
+
+        async def __aexit__(self, *a):
+            return None
+
+    with patch("src.llm.baseline._avg_goals", fake_avg), \
+         patch("src.llm.baseline.AsyncSessionLocal") as SLC:
+        SLC.return_value = FakeCM()
+
+        result = await predict_baseline(1)
+
+    assert isinstance(result, PredictResult)
+    assert result.mode == "baseline"
+    assert result.provider == "baseline"
+    assert result.model == "baseline"
+    assert result.prompt_version == "baseline_v1"
+    assert result.pred_home_goals == 2.0
+    assert result.pred_away_goals == 2.0
+    assert result.pred_1x2 == "X"
+    assert result.subjective_confidence == 0.5
+    assert result.prompt_tokens == 0
+    assert result.completion_tokens == 0
+    assert result.latency_ms == 0
+    assert result.status == "success"
+    assert "非投注建议" in (result.reasoning or "")
+    # baseline 不构建 LLM 上下文,但字段必须存在且可安全序列化
+    assert result.context == ""
+    # 原始统计快照保留在 raw 中
+    assert result.raw is not None
+    assert "home_avg" in result.raw
+
+
+# ============================================================
+# 2. MultiPredictResult 与扩展后的 PredictResult 兼容
+# ============================================================
+
+
+def test_multi_predict_result_is_predict_result_alias():
+    """multi 结果不再是重复定义的 dataclass,而是扩展 PredictResult 的别名。"""
+    assert MultiPredictResult is PredictResult
+
+
+def test_multi_result_constructor_kwargs_still_supported():
+    """orchestrator 现有构造调用的全部字段 kwargs 必须仍可构造(别名完整性)。"""
+    # 与 orchestrator.predict_match_multi 的 return MultiPredictResult(...) 逐一对应
+    result = MultiPredictResult(
+        prediction_id=1,
+        provider="openai",
+        model="gpt-x",
+        prompt_version="multi_v1",
+        mode="multi",
+        pred_home_goals=2.0,
+        pred_away_goals=1.0,
+        alt_pred_home_goals=None,
+        alt_pred_away_goals=None,
+        pred_1x2="1",
+        subjective_confidence=0.7,
+        reasoning="r",
+        status="success",
+        agent_outputs=[{"agent": "form"}],
+        agent_weights={"form": 0.2},
+        context="ctx",
+        latency_ms=100,
+        prompt_tokens=10,
+        completion_tokens=5,
+        raw={"final": True},
+    )
+    assert result.mode == "multi"
+    assert result.agent_outputs == [{"agent": "form"}]
+    assert result.agent_weights == {"form": 0.2}
+    assert result.prompt_tokens == 10
+    assert result.completion_tokens == 5
+
+
+# ============================================================
+# 3. 路由去 dict 分支(源码守卫)
+# ============================================================
+
+
+def test_predict_route_has_no_dict_branch():
+    """PredictOut 映射必须统一走属性访问,禁止 isinstance(result, dict) 回潮。"""
+    src = ROUTE_PATH.read_text(encoding="utf-8")
+    assert "isinstance(result, dict)" not in src
+    assert ".get(\"pred_home_goals\")" not in src
+
+
+# ============================================================
+# 4. _persist_baseline 属性映射(baseline 落库语义不变)
+# ============================================================
+
+
+class _FakeUoW:
+    """替代 get_uow 的最小上下文管理器。"""
+
+    def __init__(self):
+        self.session = SimpleNamespace()
+
+    async def __aenter__(self):
+        return self.session
+
+    async def __aexit__(self, *a):
+        return None
+
+
+@pytest.mark.asyncio
+async def test_persist_baseline_maps_attributes(monkeypatch):
+    captured = {}
+
+    async def fake_upsert(session, **kwargs):
+        captured.update(kwargs)
+        return SimpleNamespace(id=77)
+
+    monkeypatch.setattr("src.db.unit_of_work.get_uow", lambda: _FakeUoW())
+    monkeypatch.setattr("src.llm.predict._upsert_prediction", fake_upsert)
+
+    from src.api.routes.predict import _persist_baseline
+
+    baseline = PredictResult(
+        prediction_id=0,  # baseline 不在服务层落库,由 _persist_baseline 落库后取得真实 id
+        provider="baseline",
+        model="baseline",
+        prompt_version="baseline_v1",
+        mode="baseline",
+        pred_home_goals=2.0,
+        pred_away_goals=1.0,
+        alt_pred_home_goals=None,
+        alt_pred_away_goals=None,
+        pred_1x2="1",
+        subjective_confidence=0.5,
+        reasoning="r",
+        context="",
+        status="success",
+        latency_ms=0,
+        prompt_tokens=0,
+        completion_tokens=0,
+        raw={"home_avg": 2.1, "away_avg": 1.4},
+    )
+
+    pid = await _persist_baseline(1, baseline)
+
+    assert pid == 77
+    assert captured["match_id"] == 1
+    assert captured["provider_name"] == "baseline"
+    assert captured["model"] == "baseline"
+    assert captured["mode"] == "baseline"
+    assert captured["run_type"] == "live"
+    v = captured["values"]
+    assert v["prompt_version"] == "baseline_v1"
+    assert v["pred_home_goals"] == 2.0
+    assert v["pred_away_goals"] == 1.0
+    assert v["pred_1x2"] == "1"
+    assert v["subjective_confidence"] == 0.5
+    assert v["prompt_tokens"] == 0
+    assert v["completion_tokens"] == 0
+    assert v["latency_ms"] == 0
+    assert v["raw_response"] == {"home_avg": 2.1, "away_avg": 1.4}
+    assert v["status"] == "success"