Shadow Admission 阶段详细文档
文档ID: QSP-SA-v1.0 阶段编号: 06 阶段名称: Shadow Admission (影子准入) 日期: 2026-03-26 状态: v1.0 负责角色: PM + Risk + Quant Researcher + Trader
目录
- 阶段定义与核心目的
- 为什么需要 Shadow Admission
- Shadow Admission 与实盘的关系
- 执行语义补充
- 撮合假设明确
- 容量评估
- 成本模型精化
- 监控方案设计
- 风险控制机制
- Formal Gate 要求
- 常见错误与防范
- 输出 Artifact
- 投产交接
- 影子准入文档模板
1. 阶段定义与核心目的
1.1 阶段定义
Shadow Admission (影子准入) 是策略进入模拟实盘环境前的治理准入阶段。
核心特征：
- 模拟实盘: 在接近真实的环境中运行策略
- 治理准入: 不等于最终投产，只是更高级别的准入
- 风险隔离: 资金隔离，不会影响实盘
- 全面验证: 验证策略在更真实条件下的表现
1.2 核心目的
为什么需要 Shadow Admission：
| 目的 | 说明 | 价值 |
|---|---|---|
| 执行验证 | 验证策略在真实执行环境下是否可行 | 发现回测未暴露的问题 |
| 容量验证 | 验证策略在实际资金规模下是否可行 | 避免容量不足导致收益衰减 |
| 成本验证 | 验证成本模型是否准确 | 避免成本吞噬利润 |
| 风险验证 | 验证风险控制机制是否有效 | 防止极端损失 |
| 监控验证 | 验证监控方案是否完善 | 确保投产后的可观测性 |
1.3 与前序阶段的关系
Holdout Validation (最终验证)
↓ PASS
Shadow Admission (影子准入)
↓ PASS
Production (投产)
关键差异：
stage_differences:
backtest_ready:
environment: "历史数据回测"
execution: "模拟撮合"
cost: "简化成本模型"
focus: "策略逻辑验证"
shadow_admission:
environment: "模拟实盘 (实时数据)"
execution: "模拟撮合或小规模实盘"
cost: "精确成本模型"
focus: "执行可行性验证"
production:
environment: "实盘"
execution: "真实撮合"
cost: "真实成本"
focus: "策略运行"2. 为什么需要 Shadow Admission
2.1 回测与实盘的差距
常见差距：
gaps_between_backtest_and_production:
execution_gaps:
issue: "撮合假设过于理想"
examples:
- "假设立即成交"
- "假设无滑点"
- "假设无流动性约束"
consequence: "实盘收益大幅低于回测"
cost_gaps:
issue: "成本模型不准确"
examples:
- "低估手续费"
- "忽略融资成本"
- "忽略税务成本"
consequence: "净利润被成本吞噬"
capacity_gaps:
issue: "容量估计不足"
examples:
- "忽略自冲击"
- "忽略流动性约束"
- "忽略市场影响"
consequence: "规模扩大后收益衰减"
operational_gaps:
issue: "运营问题未考虑"
examples:
- "系统故障"
- "网络延迟"
- "人为错误"
consequence: "策略无法正常运行"2.2 Shadow Admission 的价值
价值示例：
example_value:
case_1_execution_issue:
backtest_result: "Sharpe 2.5, 年化 30%"
shadow_discovery: "订单无法及时成交，实际滑点 0.5%"
shadow_result: "Sharpe 1.2, 年化 12%"
action: "优化订单逻辑或放弃策略"
case_2_capacity_issue:
backtest_result: "$1M 资金下 Sharpe 2.0"
shadow_discovery: "$500K 后自冲击显著"
shadow_result: "$1M 资金下 Sharpe 0.8"
action: "限制规模或修改策略"
case_3_cost_issue:
backtest_result: "毛利 25%, 净利 20%"
shadow_discovery: "实际成本 8%, 净利 17%"
shadow_result: "仍可接受，但收益低于预期"
action: "优化成本或调整预期"3. Shadow Admission 与实盘的关系
3.1 模拟实盘 vs 真实实盘
对比表：
shadow_vs_production:
capital:
shadow: "隔离资金，不影响实盘"
production: "真实资金"
execution:
shadow: "模拟撮合 (或小规模真实撮合)"
production: "真实撮合"
risk:
shadow: "策略风险不影响实际资产"
production: "策略风险影响实际资产"
duration:
shadow: "通常 1-3 个月"
production: "长期运行"
purpose:
shadow: "验证策略可行性"
production: "实现策略收益"3.2 Shadow Admission 的两种模式
模式 A：纯模拟模式：
pure_simulation_mode:
execution: "模拟撮合引擎"
data_feed: "实时市场数据"
capital: "虚拟资金"
orders: "生成订单但不实际执行"
fills: "基于市场数据模拟成交"
advantages:
- "零资金风险"
- "快速迭代"
- "可测试极端情况"
disadvantages:
- "撮合假设可能仍不真实"
- "无法验证真实执行"模式 B：小规模实盘模式：
small_scale_production_mode:
execution: "真实撮合"
data_feed: "实时市场数据"
capital: "小规模真实资金 (如 $10K)"
orders: "真实订单执行"
fills: "真实成交"
advantages:
- "完全真实的执行环境"
- "验证真实成本"
disadvantages:
- "有资金风险"
- "可能影响市场 (如资金过小)"3.3 模式选择
选择指南：
mode_selection_criteria:
use_pure_simulation:
- "新策略首次验证"
- "执行逻辑复杂，需要充分测试"
- "资金预算有限"
- "市场流动性充足"
use_small_scale_production:
- "策略已经过充分验证"
- "需要验证真实撮合"
- "需要验证真实成本"
- "有足够的风险承受能力"4. 执行语义补充
4.1 执行语义定义
执行语义 是策略如何生成和执行订单的详细规范。
必需内容：
execution_semantics:
order_generation:
- "订单生成时机"
- "订单类型 (市价/限价)"
- "订单大小计算"
- "订单方向 (多头/空头)"
order_routing:
- "交易所选择"
- "路由策略"
- "拆单规则"
execution_logic:
- "成交逻辑"
- "部分成交处理"
- "成交失败处理"
position_management:
- "持仓管理"
- "仓位调整"
- "平仓逻辑"4.2 执行语义文档模板
详细模板：
## 执行语义规范
### 信号到订单的转换
**信号读取**:
- 信号源: {信号来源}
- 读取频率: {频率}
- 信号阈值: {阈值}
**订单生成**:
```python
def generate_order(signal, current_position):
"""
根据信号生成订单
"""
if signal > entry_threshold and current_position == 0:
return Order(type='MARKET', side='BUY', size=calculate_size())
elif signal < exit_threshold and current_position > 0:
return Order(type='MARKET', side='SELL', size=current_position)
else:
return None订单类型选择
| 场景 | 订单类型 | 理由 |
|---|---|---|
| 入场 | {类型} | {理由} |
| 出场 | {类型} | {理由} |
| 止损 | {类型} | {理由} |
订单大小计算
基础订单大小:
size = capital * leverage * target_exposure / current_price
约束条件:
- 最小订单: {约束}
- 最大订单: {约束}
- 流动性约束: {约束}
部分成交处理
策略:
- 订单拆分: {是否}
- 拆分规则: {规则}
- 超时处理: {处理方式}
成交失败处理
重试逻辑:
- 重试次数: {N}
- 重试间隔: {时间}
- 失败处理: {处理方式}
### 4.3 执行语义验证
**验证检查**：
```python
def validate_execution_semantics(execution_spec):
"""
验证执行语义的完整性
"""
checks = {
'order_generation_defined': 'order_generation' in execution_spec,
'order_routing_defined': 'order_routing' in execution_spec,
'execution_logic_defined': 'execution_logic' in execution_spec,
'position_management_defined': 'position_management' in execution_spec,
'error_handling_defined': 'error_handling' in execution_spec,
}
all_passed = all(checks.values())
return {
'passed': all_passed,
'checks': checks,
'missing': [k for k, v in checks.items() if not v]
}
5. 撮合假设明确
5.1 撮合假设的重要性
撮合假设 是关于订单如何成交的假设。
关键假设：
filling_assumptions:
market_orders:
assumption: "市价单立即成交"
reality: "可能有滑点和部分成交"
mitigation: "记录滑点，调整预期"
limit_orders:
assumption: "限价单在指定价格成交"
reality: "可能不成交"
mitigation: "设置超时，转市价"
slippage:
assumption: "滑点为固定值或百分比"
reality: "滑点随市场条件变化"
mitigation: "使用动态滑点模型"
partial_fills:
assumption: "订单完全成交"
reality: "可能部分成交"
mitigation: "处理部分成交逻辑"5.2 撮合模型
撮合模型示例：
class FillingModel:
"""
撮合模型
"""
def __init__(self, config):
self.slippage_model = config['slippage_model']
self.partial_fill_model = config['partial_fill_model']
self.fill_probability = config.get('fill_probability', 1.0)
def simulate_fill(self, order, market_data):
"""
模拟订单成交
"""
# 1. 检查是否成交
if random.random() > self.fill_probability:
return None # 未成交
# 2. 计算滑点
slippage = self.calculate_slippage(order, market_data)
# 3. 计算成交价格
if order.side == 'BUY':
fill_price = market_data['ask'] + slippage
else:
fill_price = market_data['bid'] - slippage
# 4. 计算成交量
fill_size = self.calculate_fill_size(order, market_data)
return {
'price': fill_price,
'size': fill_size,
'timestamp': market_data['timestamp']
}
def calculate_slippage(self, order, market_data):
"""
计算滑点
"""
if self.slippage_model['type'] == 'fixed':
return self.slippage_model['value']
elif self.slippage_model['type'] == 'percentage':
return order.price * self.slippage_model['value']
elif self.slippage_model['type'] == 'dynamic':
# 基于市场波动率的动态滑点
volatility = market_data['volatility']
return order.price * volatility * self.slippage_model['multiplier']
def calculate_fill_size(self, order, market_data):
"""
计算成交量
"""
if self.partial_fill_model['type'] == 'full':
return order.size
elif self.partial_fill_model['type'] == 'fraction':
# 基于流动性的部分成交
liquidity_ratio = min(
order.size / market_data['volume'],
self.partial_fill_model['max_fill_ratio']
)
return order.size * liquidity_ratio5.3 撮合假设文档
文档模板：
filling_assumptions_spec:
market_orders:
fill_probability: 1.0 # 100% 成交
slippage_model:
type: "dynamic"
base_bps: 5 # 基础 5bp
volatility_multiplier: 0.1 # 波动率系数
limit_orders:
fill_probability: 0.8 # 80% 成交
time_in_force: "GTC" # Good Till Cancelled
expiry_minutes: 60
partial_fills:
enabled: true
max_fill_ratio: 0.2 # 最多成交 20% 的量
min_fill_size: 0.001 # 最小成交量
market_impact:
enabled: true
model: "square_root"
impact_coefficient: 0.01
documentation: >
撮合假设基于历史订单统计分析。
在 Shadow Admission 阶段将验证这些假设的准确性。6. 容量评估
6.1 容量的定义
策略容量 是策略能够容纳的资金规模上限。
容量限制因素：
capacity_limiting_factors:
liquidity:
factor: "市场流动性"
description: "日交易量限制可交易规模"
measure: "策略交易量 / 市场交易量"
market_impact:
factor: "自冲击"
description: "策略自身交易对价格的影响"
measure: "冲击成本"
concentration:
factor: "持仓集中度"
description: "单一标的上过大仓位"
measure: "单一持仓 / 总资金"
execution:
factor: "执行能力"
description: "订单执行速度和能力"
measure: "订单成交率和延迟"6.2 容量评估方法
评估方法：
def estimate_strategy_capacity(backtest_results, market_data):
"""
估计策略容量
"""
capacity_analysis = {}
# 1. 流动性约束
# 策略日均交易量不应超过市场日均交易量的 1%
avg_strategy_volume = backtest_results['daily_volume'].mean()
avg_market_volume = market_data['daily_volume'].mean()
liquidity_capacity = avg_market_volume * 0.01 / backtest_results['turnover'].mean()
# 2. 自冲击约束
# 自冲击成本 = coefficient * sqrt(策略交易量 / 市场交易量)
# 设定最大可接受冲击成本为 5bp
max_impact_cost = 0.0005 # 5bp
impact_coefficient = 0.01
# 反推最大交易量
max_volume_ratio = (max_impact_cost / impact_coefficient) ** 2
impact_capacity = avg_market_volume * max_volume_ratio
# 3. 集中度约束
# 单一持仓不超过总资金的 10%
max_single_position = backtest_results['max_position_size'].max()
concentration_capacity = max_single_position / 0.10
# 4. 综合容量
capacity_analysis['liquidity_capacity'] = liquidity_capacity
capacity_analysis['impact_capacity'] = impact_capacity
capacity_analysis['concentration_capacity'] = concentration_capacity
# 取最小值作为策略容量
capacity_analysis['estimated_capacity'] = min(
liquidity_capacity,
impact_capacity,
concentration_capacity
)
return capacity_analysis6.3 容量衰减曲线
容量与收益的关系：
def simulate_capacity_decay(backtest_results, capital_range):
"""
模拟不同资金规模下的收益衰减
"""
decay_curve = []
for capital in capital_range:
# 按比例放大交易量
scaled_volume = backtest_results['volume'] * (capital / backtest_results['base_capital'])
# 计算冲击成本
impact_cost = calculate_impact_cost(scaled_volume, backtest_results['market_volume'])
# 计算净收益 (扣除冲击成本)
gross_return = backtest_results['return']
net_return = gross_return - impact_cost
# 计算 Sharpe
sharpe = net_return.mean() / net_return.std() * np.sqrt(252)
decay_curve.append({
'capital': capital,
'gross_return': gross_return.mean() * 252,
'impact_cost': impact_cost.mean() * 252,
'net_return': net_return.mean() * 252,
'sharpe': sharpe
})
return pd.DataFrame(decay_curve)6.4 容量评估报告
报告模板：
## 容量评估报告
### 估计容量
| 约束类型 | 容量上限 | 限制因素 |
|---------|---------|---------|
| 流动性约束 | ${X}M | 日均交易量 |
| 自冲击约束 | ${Y}M | 冲击成本 |
| 集中度约束 | ${Z}M | 单一持仓限制 |
**综合容量估计**: ${min(X,Y,Z)}M
### 容量衰减曲线
| 资金规模 | 毛收益 | 冲击成本 | 净收益 | Sharpe |
|---------|--------|---------|--------|--------|
| $100K | XX% | X% | XX% | X.XX |
| $500K | XX% | X% | XX% | X.XX |
| $1M | XX% | X% | XX% | X.XX |
| $5M | XX% | X% | XX% | X.XX |
### 推荐规模
**保守规模**: ${容量 * 0.5}M
**正常规模**: ${容量 * 0.7}M
**最大规模**: ${容量}M
### 风险提示
{容量相关的风险提示}7. 成本模型精化
7.1 成本组成
完整成本模型：
cost_components:
trading_costs:
commission:
description: "交易手续费"
model: "按交易量百分比"
typical_value: "0.1% (双向)"
slippage:
description: "滑点成本"
model: "动态模型"
typical_value: "2-10bp"
market_impact:
description: "自冲击成本"
model: "平方根模型"
typical_value: "与交易量相关"
holding_costs:
funding_cost:
description: "资金成本"
model: "利率 × 持仓价值"
borrowing_cost:
description: "借券/做空成本"
model: "费率 × 借券价值"
operational_costs:
infrastructure:
description: "基础设施成本"
model: "固定成本"
monitoring:
description: "监控和维护成本"
model: "固定成本"
other_costs:
taxes:
description: "税务成本"
model: "根据税率计算"
regulatory:
description: "监管成本"
model: "固定成本"7.2 成本模型实现
成本计算函数：
def calculate_total_costs(trades, positions, cost_config):
"""
计算总成本
"""
total_costs = 0
# 1. 交易成本
for trade in trades:
# 手续费
commission = trade['value'] * cost_config['commission_rate']
total_costs += commission
# 滑点
slippage = trade['value'] * cost_config['slippage_rate']
total_costs += slippage
# 市场冲击
market_impact = cost_config['impact_coefficient'] * np.sqrt(
trade['value'] / trade['market_volume']
) * trade['value']
total_costs += market_impact
# 2. 持仓成本
for position in positions:
# 资金成本
funding_cost = position['value'] * cost_config['funding_rate'] * position['days_held']
total_costs += funding_cost
# 借券成本 (如果有空头)
if position['size'] < 0:
borrowing_cost = abs(position['value']) * cost_config['borrowing_rate'] * position['days_held']
total_costs += borrowing_cost
return total_costs7.3 成本模型验证
验证方法：
cost_model_validation:
historical_comparison:
method: "与历史实际成本对比"
requirement: "模型成本 vs 实际成本差异 < 20%"
shadow_admission_testing:
method: "在影子准入中验证"
requirement: "记录实际成本，调整模型"
sensitivity_analysis:
method: "参数敏感性分析"
requirement: "关键参数变动 ±50% 评估影响"8. 监控方案设计
8.1 监控指标体系
核心监控指标：
monitoring_metrics:
performance_metrics:
- name: "daily_pnl"
description: "日盈亏"
frequency: "daily"
- name: "cumulative_return"
description: "累计收益"
frequency: "daily"
- name: "sharpe_ratio"
description: "夏普比率"
frequency: "weekly"
- name: "max_drawdown"
description: "最大回撤"
frequency: "daily"
execution_metrics:
- name: "fill_rate"
description: "成交率"
frequency: "daily"
- name: "slippage"
description: "平均滑点"
frequency: "daily"
- name: "order_latency"
description: "订单延迟"
frequency: "hourly"
risk_metrics:
- name: "position_concentration"
description: "持仓集中度"
frequency: "daily"
- name: "var_95"
description: "95% VaR"
frequency: "daily"
- name: "beta"
description: "市场 Beta"
frequency: "weekly"
operational_metrics:
- name: "system_uptime"
description: "系统可用性"
frequency: "hourly"
- name: "data_latency"
description: "数据延迟"
frequency: "hourly"
- name: "error_rate"
description: "错误率"
frequency: "daily"8.2 告警规则
告警配置：
alert_rules:
performance_alerts:
- metric: "daily_pnl"
condition: "< -$10,000"
severity: "HIGH"
action: "立即通知 PM 和 Risk"
- metric: "max_drawdown"
condition: "< -15%"
severity: "MEDIUM"
action: "通知团队，评估是否暂停"
execution_alerts:
- metric: "fill_rate"
condition: "< 80%"
severity: "HIGH"
action: "检查撮合问题"
- metric: "slippage"
condition: "> 20bp"
severity: "MEDIUM"
action: "评估订单策略"
risk_alerts:
- metric: "position_concentration"
condition: "> 20%"
severity: "MEDIUM"
action: "评估集中度风险"
- metric: "var_95"
condition: ">$50,000"
severity: "HIGH"
action: "降低仓位"
operational_alerts:
- metric: "system_uptime"
condition: "< 99%"
severity: "HIGH"
action: "立即修复"
- metric: "error_rate"
condition: "> 5%"
severity: "MEDIUM"
action: "调查错误原因"8.3 监控仪表盘
仪表盘设计：
dashboard_layout:
overview_section:
- "累计收益曲线"
- "当前持仓"
- "今日盈亏"
- "关键指标"
performance_section:
- "收益分布"
- "回撤分析"
- "Sharpe 趋势"
execution_section:
- "成交率"
- "滑点分析"
- "订单延迟"
risk_section:
- "持仓分析"
- "VaR 分析"
- "Beta 分析"
operational_section:
- "系统状态"
- "数据质量"
- "错误日志"9. 风险控制机制
9.1 风险限制
风险限制设置：
risk_limits:
position_limits:
max_total_exposure: "1.0" # 100% 总资金
max_single_position: "0.10" # 10% 单一持仓
max_sector_exposure: "0.30" # 30% 行业暴露
leverage_limits:
max_gross_exposure: "1.5" # 150% 总敞口
max_net_exposure: "0.5" # 50% 净敞口
drawdown_limits:
max_daily_drawdown: "-0.05" # -5% 日最大回撤
max_total_drawdown: "-0.15" # -15% 总最大回撤"
loss_limits:
max_daily_loss: "-$20,000" # 日最大亏损
max_weekly_loss: "-$50,000" # 周最大亏损"
concentration_limits:
max_correlation_positions: "5" # 最大相关持仓数
min_diversification_ratio: "0.5" # 最小分散度9.2 风险控制流程
风险控制机制：
class RiskController:
"""
风险控制器
"""
def __init__(self, risk_limits):
self.limits = risk_limits
self.current_state = {}
def check_order(self, order, current_positions):
"""
检查订单是否符合风险限制
"""
checks = {}
# 1. 检查单一持仓限制
if order.symbol in current_positions:
new_size = current_positions[order.symbol] + order.size
checks['single_position'] = abs(new_size) <= self.limits['max_single_position']
else:
checks['single_position'] = abs(order.size) <= self.limits['max_single_position']
# 2. 检查总敞口限制
total_exposure = sum(abs(p) for p in current_positions.values())
new_exposure = total_exposure + abs(order.size)
checks['total_exposure'] = new_exposure <= self.limits['max_total_exposure']
# 3. 检查净敞口限制
net_exposure = sum(current_positions.values())
new_net = net_exposure + order.size
checks['net_exposure'] = abs(new_net) <= self.limits['max_net_exposure']
# 4. 检查回撤限制
checks['drawdown'] = self.check_drawdown_limit()
return all(checks.values()), checks
def check_drawdown_limit(self):
"""
检查回撤限制
"""
current_drawdown = self.calculate_current_drawdown()
return current_drawdown >= self.limits['max_total_drawdown']
def check_loss_limit(self, daily_pnl):
"""
检查亏损限制
"""
return daily_pnl >= self.limits['max_daily_loss']9.3 紧急停止机制
紧急停止触发条件：
emergency_stop_triggers:
immediate_stop:
- "系统故障或数据异常"
- "超出最大回撤限制"
- "超出最大亏损限制"
- "风险控制失效"
gradual_stop:
- "持续亏损 (如连续 5 天)"
- "市场环境变化"
- "策略性能下降"
stop_procedure:
1. "停止新订单"
2. "平仓现有持仓"
3. "记录停止原因"
4. "通知相关人员"
5. "分析停止原因"10. Formal Gate 要求
10.1 Shadow Admission 阶段 Formal Gate
必需检查项：
shadow_admission_formal_gates:
execution_semantics_complete:
requirement: "执行语义已完整定义"
criteria:
- "订单生成逻辑清晰"
- "撮合假设明确"
- "异常处理完善"
evidence: "execution_semantics.md"
status: "PASS | FAIL"
capacity_assessed:
requirement: "策略容量已评估"
criteria:
- "容量估计完成"
- "衰减曲线分析完成"
- "推荐规模明确"
evidence: "capacity_assessment_report.md"
status: "PASS | FAIL"
cost_model_refined:
requirement: "成本模型已精化"
criteria:
- "所有成本项已考虑"
- "成本模型已验证"
- "成本在可接受范围"
evidence: "cost_model_report.md"
status: "PASS | FAIL"
monitoring_designed:
requirement: "监控方案已设计"
criteria:
- "监控指标完整"
- "告警规则明确"
- "仪表盘设计完成"
evidence: "monitoring_plan.md"
status: "PASS | FAIL"
risk_controls_defined:
requirement: "风险控制机制已定义"
criteria:
- "风险限制明确"
- "风险控制流程清晰"
- "紧急停止机制完善"
evidence: "risk_control_plan.md"
status: "PASS | FAIL"
shadow_period_completed:
requirement: "影子准入期已完成"
criteria:
- "运行时间 ≥ 1 个月"
- "市场环境代表性"
- "问题已记录和解决"
evidence: "shadow_period_summary.md"
status: "PASS | FAIL"10.2 决策状态
可能的状态：
verdict_states:
PASS:
description: "所有 Formal Gate 通过，可进入投产"
conditions:
- "Shadow 期表现符合预期"
- "执行可行性确认"
- "风险可控"
- "监控方案完善"
CONDITIONAL_PASS:
description: "核心要求通过，但有条件限制"
conditions:
- "限制规模"
- "加强监控"
- "定期审查"
RETRY:
description: "执行或成本问题，修复后重试"
scope:
- "优化执行逻辑"
- "调整撮合假设"
- "修改成本模型"
EXTEND_SHADOW:
description: "需要延长影子准入期"
reasons:
- "市场环境不典型"
- "需要更多数据验证"
- "发现新问题需要解决"
NO_GO:
description: "策略不适合实盘，终止"
reasons:
- "执行不可行"
- "容量不足"
- "成本过高"
- "风险不可控"11. 常见错误与防范
11.1 错误清单
| 错误类型 | 描述 | 后果 | 防范措施 |
|---|---|---|---|
| 过度乐观 | 假设撮合过于理想 | 实盘收益远低于预期 | 使用保守撮合假设 |
| 忽略容量 | 不评估策略容量 | 规模扩大后收益衰减 | 提前评估容量 |
| 低估成本 | 成本模型不完整 | 净利润被成本吞噬 | 考虑所有成本项 |
| 缺少监控 | 不设计监控方案 | 问题发现不及时 | 完善监控体系 |
| 风险不足 | 风险控制不严格 | 极端情况下大额亏损 | 严格风险限制 |
| 急于投产 | Shadow 期不够充分 | 实盘暴露未发现问题 | 充分运行 Shadow |
11.2 防范机制
流程层面：
safety_mechanisms:
conservative_assumptions:
principle: "使用保守假设，给实盘留余量"
application:
- "撮合假设保守"
- "成本估计高估"
- "容量估计低估"
staged_rollout:
principle: "分阶段投产，逐步扩大规模"
stages:
- "纯模拟"
- "小规模实盘"
- "中等规模"
- "全规模"
continuous_validation:
principle: "持续验证假设"
actions:
- "对比实际与假设"
- "调整模型"
- "更新预期"
independent_review:
principle: "独立审查关键决策"
reviewers:
- "Risk 独立审查风险限制"
- "Trader 独立审查执行方案"
- "PM 独立审查整体可行性"12. 输出 Artifact
12.1 机器可读产物
必需文件：
machine_readable_artifacts:
execution_semantics.yaml:
description: "执行语义规范"
content:
- "订单生成逻辑"
- "撮合假设"
- "异常处理"
capacity_assessment.yaml:
description: "容量评估结果"
content:
- "容量估计"
- "衰减曲线"
- "推荐规模"
cost_model.yaml:
description: "成本模型配置"
content:
- "成本项定义"
- "成本参数"
- "验证结果"
monitoring_config.yaml:
description: "监控配置"
content:
- "监控指标"
- "告警规则"
- "仪表盘配置"
risk_limits.yaml:
description: "风险限制配置"
content:
- "风险限制"
- "控制流程"
- "停止机制"
shadow_performance.parquet:
description: "影子期表现数据"
content:
- "日收益数据"
- "执行数据"
- "成本数据"12.2 人类可读产物
必需文档：
human_readable_artifacts:
shadow_admission_report.md:
description: "影子准入总结报告"
sections:
- "执行摘要"
- "执行语义验证"
- "容量评估"
- "成本模型验证"
- "监控方案验证"
- "风险控制验证"
- "Shadow 期表现"
- "问题与解决方案"
- "投产建议"
execution_semantics_documentation.md:
description: "执行语义详细文档"
content:
- "订单生成详细逻辑"
- "撮合假设详细说明"
- "异常处理详细流程"
monitoring_plan.md:
description: "监控方案文档"
content:
- "监控指标定义"
- "告警规则说明"
- "仪表盘使用指南"
risk_control_plan.md:
description: "风险控制方案文档"
content:
- "风险限制说明"
- "控制流程详细说明"
- "紧急停止流程"
shadow_period_summary.md:
description: "影子期总结"
content:
- "运行期间关键事件"
- "性能表现总结"
- "发现的问题"
- "实施的调整"13. 投产交接
13.1 交接内容
Shadow Admission → Production 交接清单：
production_handover:
execution_ready:
execution_semantics: "已验证的执行语义"
filling_model: "已验证的撮合模型"
infrastructure: "就绪的基础设施"
capacity_confirmed:
recommended_scale: "推荐规模"
max_safe_scale: "最大安全规模"
scale_up_plan: "扩容计划"
cost_confirmed:
cost_model: "验证后的成本模型"
expected_net_return: "预期净收益"
cost_monitoring: "成本监控方案"
monitoring_ready:
monitoring_system: "已部署的监控系统"
alert_system: "已配置的告警系统"
dashboard: "已配置的仪表盘"
risk_controls_ready:
risk_limits: "已配置的风险限制"
risk_controller: "已部署的风险控制器"
emergency_stop: "已测试的紧急停止"
documentation_complete:
all_plans: "所有方案文档完整"
runbooks: "操作手册完整"
contact_list: "联系人清单"13.2 投产前最终检查
最终检查清单：
pre_production_checks:
technical_readiness:
- [ ] 系统已部署并通过测试
- [ ] 监控已配置并测试
- [ ] 告警已配置并测试
- [ ] 风险控制已部署并测试
- [ ] 紧急停止已测试
operational_readiness:
- [ ] 操作员已培训
- [ ] 值班表已安排
- [ ] 联系人已确认
- [ ] 应急流程已明确
governance_readiness:
- [ ] PM 批准投产
- [ ] Risk 批准风险限制
- [ ] Trader 确认执行方案
- [ ] 合规审查通过
documentation_readiness:
- [ ] 所有文档完整
- [ ] 操作手册更新
- [ ] 应急预案准备
- [ ] 交接记录完整13.3 投产决策
决策框架：
production_decision:
go_live:
conditions:
- "所有 Formal Gate 通过"
- "Shadow 期表现良好"
- "无未解决的重大问题"
- "所有准备工作完成"
conditional_go_live:
conditions:
- "核心 Formal Gate 通过"
- "Shadow 期表现可接受"
- "有已知但有条件的问题"
- "准备工作基本完成"
conditions_list:
- "限制规模在推荐范围内"
- "加强监控频率"
- "定期审查 (如每周)"
- "准备降级方案"
delay_production:
conditions:
- "有未解决的重大问题"
- "Shadow 期表现不理想"
- "准备工作未完成"
- "需要更多验证时间"
cancel_production:
conditions:
- "策略不适合实盘"
- "风险不可控"
- "成本过高"
- "执行不可行"14. 影子准入文档模板
14.1 Shadow Admission 报告模板
---
doc_id: SA-{lineage_id}-{run_id}
title: Shadow Admission Report — {策略名称}
date: YYYY-MM-DD
status: PASS | CONDITIONAL_PASS | RETRY | EXTEND_SHADOW | NO_GO
version: 1.0
owner: PM + Risk + Quant Researcher
lineage_id: {lineage_id}
run_id: {run_id}
---
## 执行摘要
**策略名称**: {策略名称}
**影子期**: {开始日期} 至 {结束日期}
**运行模式**: {纯模拟 / 小规模实盘}
**总体评估**: {PASS / CONDITIONAL_PASS / RETRY / EXTEND_SHADOW / NO_GO}
**核心发现**:
- 执行可行性: {评估结果}
- 容量评估: {容量估计}
- 成本验证: {成本与预期对比}
- 风险控制: {风险事件和应对}
**投产建议**: {建议}
## 执行语义验证
### 订单生成
**验证结果**: {PASS / FAIL}
**发现的问题**:
- {问题描述}
- {解决方案}
### 撮合假设
**假设 vs 实际**:
| 假设项 | 假设值 | 实际值 | 差异 | 状态 |
|--------|--------|--------|------|------|
| 成交率 | {假设} | {实际} | {差异} | {状态} |
| 滑点 | {假设} | {实际} | {差异} | {状态} |
**假设调整**:
- {需要调整的假设}
- {调整原因}
- {新假设值}
### 异常处理
**测试的异常情况**:
- {异常情况1}: {处理结果}
- {异常情况2}: {处理结果}
## 容量评估
### 估计容量
| 约束类型 | 容量上限 | 限制因素 |
|---------|---------|---------|
| 流动性约束 | {容量} | {因素} |
| 自冲击约束 | {容量} | {因素} |
| 集中度约束 | {容量} | {因素} |
**综合容量**: ${容量}M
### 容量衰减
| 资金规模 | 毛收益 | 冲击成本 | 净收益 | Sharpe |
|---------|--------|---------|--------|--------|
| {规模1} | {收益} | {成本} | {净收益} | {Sharpe} |
| {规模2} | {收益} | {成本} | {净收益} | {Sharpe} |
### 推荐规模
- **保守**: ${规模}M
- **正常**: ${规模}M
- **最大**: ${规模}M
## 成本模型验证
### 成本对比
| 成本项 | 回测假设 | 影子期实际 | 差异 | 状态 |
|--------|---------|-----------|------|------|
| 手续费 | {假设} | {实际} | {差异} | {状态} |
| 滑点 | {假设} | {实际} | {差异} | {状态} |
| 冲击成本 | {假设} | {实际} | {差异} | {状态} |
### 收益影响
| 指标 | 回测 | 影子期 | 差异 |
|------|------|--------|------|
| 毛收益 | {值} | {值} | {差异} |
| 总成本 | {值} | {值} | {差异} |
| 净收益 | {值} | {值} | {差异} |
| Sharpe | {值} | {值} | {差异} |
## 监控方案验证
### 监控指标
**指标表现**:
- {指标1}: {描述}
- {指标2}: {描述}
### 告警触发
**触发的告警**:
- {告警1}: {次数} 次，原因: {原因}
- {告警2}: {次数} 次，原因: {原因}
### 仪表盘
**仪表盘使用情况**:
- {使用情况}
- {改进建议}
## 风险控制验证
### 风险事件
**记录的风险事件**:
| 日期 | 事件类型 | 描述 | 应对 | 结果 |
|------|---------|------|------|------|
| {日期} | {类型} | {描述} | {应对} | {结果} |
### 风险限制
**限制测试**:
| 限制类型 | 限制值 | 是否触发 | 应对 |
|---------|--------|---------|------|
| {限制1} | {值} | {是/否} | {应对} |
| {限制2} | {值} | {是/否} | {应对} |
## Shadow 期表现
### 收益表现
| 指标 | 回测 | 影子期 | 差异 |
|------|------|--------|------|
| 累计收益 | {值} | {值} | {差异} |
| 年化收益 | {值} | {值} | {差异} |
| Sharpe | {值} | {值} | {差异} |
| 最大回撤 | {值} | {值} | {差异} |
### 执行表现
| 指标 | 目标 | 实际 | 状态 |
|------|------|------|------|
| 成交率 | {目标} | {实际} | {状态} |
| 平均滑点 | {目标} | {实际} | {状态} |
| 订单延迟 | {目标} | {实际} | {状态} |
### 市场环境
**影子期市场环境**:
- {环境描述}
- {对策略的影响}
## 发现的问题与解决方案
### 严重问题
{如果有严重问题}
### 轻微问题
{如果有轻微问题}
### 实施的调整
{实施的调整和效果}
## 投产建议
### 决策
**建议**: {GO / CONDITIONAL_GO / DELAY / CANCEL}
### 理由
{详细理由}
### 投产条件
{如果是条件投产，列出条件}
### 风险提示
{投产风险提示}
### 后续计划
{投产后的监控和调整计划}
## 附录
### A. 详细执行数据
{详细执行数据}
### B. 成本明细
{成本明细}
### C. 监控图表
{监控图表}
### D. 风险事件日志
{风险事件日志}文档版本: v1.0 最后更新: 2026-03-26 下次评审: 2026-06-26