diff --git a/docs/adaptive_allocation_control_plane.zh-CN.md b/docs/adaptive_allocation_control_plane.zh-CN.md
index 39c7bc9..7f8dd82 100644
--- a/docs/adaptive_allocation_control_plane.zh-CN.md
+++ b/docs/adaptive_allocation_control_plane.zh-CN.md
@@ -32,7 +32,8 @@ quant-adaptive-selection --input selection-input.json --output selection-decisio
 - 不读取新闻叙事并直接交易；因子必须来自版本化的数据链。
 - 数据过期、状态不明、置信度不足、插件未批准、平台未对账或不健康时 fail-closed。
 - 插件只能给出 `0..1` 的风险缩放，不能增加风险或提交订单。
-- 只有已获 Shadow 准入、并绑定 immutable release 的候选可以被排名。
+- 只有已获 Shadow 准入、并绑定 immutable release 的候选可以被排名；`shadow_candidate`
+  仅可得到“建议进入 Shadow”的零仓位结论，不能由此启动 runtime。
 - P0 不能改变平台、策略、插件挂载、仓位或调度器。
 
 后续 M1 才能把 Shadow 结论呈现给人工；M2/M3 还必须经过独立的 Paper、Canary、
diff --git a/src/quant_platform_kit/adaptive_allocation/contracts.py b/src/quant_platform_kit/adaptive_allocation/contracts.py
index 46d2ade..810ae98 100644
--- a/src/quant_platform_kit/adaptive_allocation/contracts.py
+++ b/src/quant_platform_kit/adaptive_allocation/contracts.py
@@ -14,7 +14,9 @@
 PLATFORM_HEALTH_SCHEMA = "qsl.platform_health_snapshot.v1"
 SELECTION_DECISION_SCHEMA = "qsl.selection_decision.v1"
 SHADOW_ONLY_AUTHORITY = "shadow_only"
-_SHADOW_ELIGIBLE_STAGES = frozenset({"shadow_active", "paper_active", "live_candidate", "live_enabled"})
+_SHADOW_ELIGIBLE_STAGES = frozenset(
+    {"shadow_candidate", "shadow_active", "paper_active", "live_candidate", "live_enabled"}
+)
 
 
 def _nonblank(value: str, field_name: str) -> str:
diff --git a/src/quant_platform_kit/adaptive_allocation/selector.py b/src/quant_platform_kit/adaptive_allocation/selector.py
index 14a733b..5979741 100644
--- a/src/quant_platform_kit/adaptive_allocation/selector.py
+++ b/src/quant_platform_kit/adaptive_allocation/selector.py
@@ -62,7 +62,10 @@ def select_shadow(
 
     `base_score` and `factor_exposures` must come from a separately validated,
     versioned research pipeline.  This function intentionally has no market-data
-    fetcher and does not infer financial features from narratives.
+    fetcher and does not infer financial features from narratives.  A
+    ``shadow_candidate`` that is explicitly approved for Shadow may be ranked
+    so that P0 can produce a reviewable recommendation, but the result still
+    cannot start a Shadow runtime or authorize an order.
     """
 
     health_by_platform = {item.platform_id: item for item in platform_health}
diff --git a/tests/test_adaptive_allocation.py b/tests/test_adaptive_allocation.py
index 28f05a3..31dc048 100644
--- a/tests/test_adaptive_allocation.py
+++ b/tests/test_adaptive_allocation.py
@@ -116,6 +116,21 @@ def test_shadow_selector_rejects_unhealthy_platform_or_missing_required_plugin()
     assert "no_healthy_reconciled_shadow_platform" in item.reasons
 
 
+def test_shadow_selector_can_rank_an_approved_shadow_candidate_without_starting_it():
+    decision = select_shadow(
+        decision_id="decision-004",
+        market_context=_context(),
+        candidates=[_candidate("candidate_a", 0.7, lifecycle_stage="shadow_candidate")],
+        platform_health=[_health()],
+        plugin_adjustments=[PluginRiskAdjustment("market_regime_control", 1.0)],
+        policy=_policy(),
+    )
+
+    assert decision.recommended_strategy_profile == "candidate_a"
+    assert decision.no_order is True
+    assert decision.candidate_decisions[0].proposed_weight == 0.0
+
+
 def test_plugin_risk_adjustment_cannot_increase_risk():
     with pytest.raises(ValueError, match="risk_multiplier"):
         PluginRiskAdjustment("market_regime_control", 1.01)