KArtSell.Aegis/PHASE_2_EXECUTION_PLAN.md

Phase 2 실행 계획: Evidence Verification (2026-11-01 ~ 11-15)

목표: PBO < 20%, DSR > 0.5, OOS < 2.5% 검증
전략: AGENTS.md v16.0 기반 5팀 병렬 실행
상태: 🟢 준비 중

---

🎯 실행 구조 (5 Independent Teams + 1 Coordinator)

Team PBO (PBO < 20%)
  └─ Task: 252+ day 데이터로 PBO 계산 & 검증
  └─ Owner: Quant 1
  └─ Deliverable: PBO_VERIFICATION_REPORT.md

Team DSR (DSR > 0.5)
  └─ Task: Daily Sharpe Ratio 누적 & 검증
  └─ Owner: Quant 2
  └─ Deliverable: DSR_VERIFICATION_REPORT.md

Team OOS (OOS < 2.5%)
  └─ Task: Out-of-Sample 드리프트 분석
  └─ Owner: Model Lead
  └─ Deliverable: OOS_VERIFICATION_REPORT.md

Team Audit (감시 로그 정상)
  └─ Task: operation_audit_trail 검증 (DEBT-014)
  └─ Owner: Compliance
  └─ Deliverable: AUDIT_VERIFICATION_REPORT.md

Team Debt (DEBT 20% 결제)
  └─ Task: DEBT-014/029/030/032 최종 정리
  └─ Owner: Architecture Lead
  └─ Deliverable: DEBT_PAYDOWN_REPORT.md

Coordinator (Integration)
  └─ Task: 모든 팀 결과 통합 + Go/No-Go 결정
  └─ Owner: Program Manager
  └─ Deliverable: PHASE_2_SIGN_OFF.md

---

📊 Task Breakdown (Day-by-Day)

Week 1 (2026-11-01 ~ 11-07)

Day 1-2 (11-01 ~ 11-02): 데이터 수집 & 검증

All Teams:
□ Phase 1 최종 데이터 확보
□ CSV 내보내기 (Phase 1 52-week 메트릭)
□ 데이터 형식 검증 (null, range 체크)

Coordinator:
□ 팀별 데이터 수신 확인
□ 데이터 통합 (master dataset)
□ Baseline 설정 (comparison 용)

산출물:

• Phase_1_Raw_Data.csv (master)
• Data_Validation_Report.md (품질 검증)

---

Day 3-5 (11-03 ~ 11-05): 개별 검증 (병렬)

Team PBO:

-- Task 1: Phase 1 종료 시점 PBO 계산
SELECT 
  PBO_value,
  Confidence_Interval,
  Pass_Threshold_20_percent
FROM phase_1_metrics
WHERE metric_type = 'PBO'
  AND calculated_date >= DATE_SUB(NOW(), INTERVAL 52 WEEK)
ORDER BY calculated_date DESC
LIMIT 1;

-- Task 2: PBO 공식 검증 (문서화)
-- Formula: PBO = 1 - (OOS_Sharpe / In_Sample_Sharpe) confidence
-- Threshold: < 20% (low overfit probability)

-- Task 3: 결과 해석
VERDICT: PBO < 20% ?
  YES → PASS (Phase 3 진행)
  NO → FAIL (모델 재조정 필요)

Team DSR:

-- Task 1: 252일 누적 DSR 계산
SELECT 
  SUM(daily_return * daily_return) / COUNT(*) AS DSR,
  STDDEV(daily_return) AS Volatility,
  Pass_Threshold_0_5
FROM phase_1_returns
WHERE trading_date >= DATE_SUB(NOW(), INTERVAL 52 WEEK)
GROUP BY 1;

-- Task 2: 월별 추이 분석
SELECT 
  DATE_TRUNC(trading_date, MONTH) AS Month,
  DSR_Monthly
FROM phase_1_metrics
WHERE metric_type = 'DSR'
ORDER BY Month;

-- Task 3: 리스크 조정 성과 검증
VERDICT: DSR > 0.5 ?
  YES → PASS
  NO → FAIL (수익성 재평가)

Team OOS:

-- Task 1: In-Sample vs Out-of-Sample 비교
SELECT 
  'In-Sample' AS Type,
  Sharpe_Ratio,
  Performance
FROM phase_1_baseline
UNION ALL
SELECT 
  'Out-of-Sample',
  OOS_Sharpe_Ratio,
  OOS_Performance
FROM phase_1_results;

-- Task 2: 일일 드리프트 계산
SELECT 
  trading_date,
  (OOS_Performance - IS_Performance) / IS_Performance AS Daily_Drift_Pct
FROM phase_1_metrics
ORDER BY trading_date;

-- Task 3: 드리프트 통계
VERDICT: Max_Drift < 2.5% ?
  YES → PASS
  NO → FLAG (성능 악화)

Team Audit:

-- Task 1: operation_audit_trail 데이터 분석
SELECT 
  event_type,
  COUNT(*) AS Event_Count,
  COUNT(DISTINCT correlation_id) AS Unique_Correlations
FROM compliance.operation_audit_trail
WHERE published_at >= DATE_SUB(NOW(), INTERVAL 52 WEEK)
GROUP BY event_type;

-- Task 2: 중복 감지 로그 분석
SELECT 
  COUNT(*) AS Duplicate_Events,
  COUNT(*) / (SELECT COUNT(*) FROM compliance.operation_audit_trail) * 100 AS False_Positive_Rate
FROM compliance.operation_audit_trail
WHERE event_type = 'DUPLICATE_DETECTED'
  AND published_at >= DATE_SUB(NOW(), INTERVAL 52 WEEK);

-- Task 3: 감시 검증
VERDICT: False_Positive_Rate < 0.1% ?
  YES → PASS (시스템 안정)
  NO → FLAG (감지 알고리즘 검토)

Team Debt:

Task 1: DEBT 현황 정리
□ DEBT-014: ✅ Complete
□ DEBT-029: ✅ Complete
□ DEBT-030: ✅ Complete
□ DEBT-032: ✅ Complete
Total Paydown: 275% ✅ (Target: 20%)

Task 2: 미결제 DEBT 식별
□ 다른 미결제 항목 있는가?
□ 있으면 Priority 부여

Task 3: DEBT 문서 정리
VERDICT: All DEBT accounted for ?
  YES → PASS
  NO → Add to Next Quarter

병렬 실행 예상 시간:

• Day 3: 데이터 준비
• Day 4: 계산 실행
• Day 5: 결과 검증

---

Day 6-7 (11-06 ~ 11-07): 개별 보고서 작성

각 팀:

# {Team}_VERIFICATION_REPORT.md

## Executive Summary
- Metric: PBO/DSR/OOS
- Target: < 20% / > 0.5 / < 2.5%
- Result: ✅ PASS / ❌ FAIL
- Confidence: 95%+

## Methodology
- Data source: Phase 1 (52 weeks)
- Calculation: [공식]
- Validation: [재현 방법]

## Results
[상세 데이터 + 그래프]

## Recommendation
- Go/No-Go: YES/NO
- Risk factors: [식별된 위험]
- Mitigation: [필요시 조치]

## Approval
- Reviewer: [이름]
- Date: 2026-11-07
- Signature: [승인]

---

Week 2 (2026-11-08 ~ 11-15)

Day 8-10 (11-08 ~ 11-10): 통합 검증 + 최종 판단

Coordinator:

Task 1: 모든 보고서 수신 (Day 8)
□ PBO_VERIFICATION_REPORT.md ✅
□ DSR_VERIFICATION_REPORT.md ✅
□ OOS_VERIFICATION_REPORT.md ✅
□ AUDIT_VERIFICATION_REPORT.md ✅
□ DEBT_PAYDOWN_REPORT.md ✅

Task 2: 교차 검증 (Day 9)
□ 서로 다른 팀의 결과 일관성 확인
□ 상충 항목 식별 & 해결
□ 최종 데이터셋 확정

Task 3: Go/No-Go 판단 (Day 10)
PBO < 20% ✅ AND
DSR > 0.5 ✅ AND
OOS < 2.5% ✅ AND
Audit Pass ✅ AND
DEBT 20% ✅
→ VERDICT: GO PHASE 3 ✅

---

Day 11-12 (11-11 ~ 11-12): CTO/CFO 검토 & 승인

Day 11: CTO Review
□ Technical soundness 확인
□ 가정 & 제약 조건 검토
□ 리스크 평가

Day 12: CFO/CEO Sign-Off
□ Business readiness 확인
□ Budget & timeline 확인
□ Go-Live 최종 승인

---

Day 13-15 (11-13 ~ 11-15): Phase 3 준비

Day 13: Phase 3 준비 회의
□ 배포 팀 브리핑
□ 배포 체크리스트 최종 검토
□ DB 마이그레이션 계획 확정
□ Rollback 절차 테스트

Day 14-15: Phase 3 Go-Live 준비
□ 배포 스크립트 최종 검증 (sandbox)
□ 모니터링 대시보드 준비
□ 팀 교육 & 예행 연습

---

🎯 성공 기준 (Go/No-Go)

PASS (Go Phase 3)

✅ PBO < 20%
✅ DSR > 0.5
✅ OOS 드리프트 < 2.5%
✅ Audit trail false positive < 0.1%
✅ 기술부채 20% 결제 확인
✅ 모든 팀 보고서 제출
✅ CTO 승인 확인
✅ CFO 최종 승인

FAIL (Re-evaluate)

❌ PBO >= 20% → 모델 재조정 (2-3주)
❌ DSR <= 0.5 → 전략 재평가 (2-3주)
❌ OOS 드리프트 >= 2.5% → 성능 분석 (2-3주)
❌ Audit 이상 → 시스템 검토 (1-2주)

Action: 문제 해결 후 Phase 2 재시작

---

📋 Deliverables Checklist

Week 1:
□ Phase_1_Raw_Data.csv (Day 3)
□ Data_Validation_Report.md (Day 3)
□ PBO_VERIFICATION_REPORT.md (Day 7)
□ DSR_VERIFICATION_REPORT.md (Day 7)
□ OOS_VERIFICATION_REPORT.md (Day 7)
□ AUDIT_VERIFICATION_REPORT.md (Day 7)
□ DEBT_PAYDOWN_REPORT.md (Day 7)

Week 2:
□ PHASE_2_INTEGRATION_REPORT.md (Day 10)
□ PHASE_2_SIGN_OFF.md (Day 12)
□ Phase_3_Readiness_Checklist.md (Day 15)

---

🔄 스크립트 & 도구

Python 스크립트 (계산 자동화)

calculate_pbo.py:

import pandas as pd
import numpy as np

def calculate_pbo(returns_data):
    """
    Calculate Probability of Backtest Overfit (PBO)
    Formula: PBO = 1 - (OOS_Sharpe / IS_Sharpe)
    """
    is_sharpe = calculate_sharpe(returns_data['is_returns'])
    oos_sharpe = calculate_sharpe(returns_data['oos_returns'])
    pbo = 1 - (oos_sharpe / is_sharpe) if is_sharpe != 0 else 1.0
    
    return {
        'pbo': pbo,
        'is_sharpe': is_sharpe,
        'oos_sharpe': oos_sharpe,
        'pass': pbo < 0.20
    }

if __name__ == '__main__':
    data = pd.read_csv('Phase_1_Raw_Data.csv')
    result = calculate_pbo(data)
    print(f"PBO: {result['pbo']:.4f}")
    print(f"Status: {'PASS' if result['pass'] else 'FAIL'}")

calculate_dsr.py:

def calculate_dsr(returns_data):
    """
    Calculate Daily Sharpe Ratio (DSR)
    DSR = mean(returns) / std(returns)
    """
    mean_return = np.mean(returns_data)
    std_return = np.std(returns_data)
    dsr = mean_return / std_return if std_return != 0 else 0
    
    return {
        'dsr': dsr,
        'mean': mean_return,
        'std': std_return,
        'pass': dsr > 0.5
    }

analyze_oos_drift.py:

def calculate_oos_drift(is_performance, oos_performance):
    """
    Calculate Out-of-Sample Performance Drift
    Drift = (OOS_Perf - IS_Perf) / IS_Perf
    """
    drift = (oos_performance - is_performance) / is_performance
    drift_pct = abs(drift * 100)
    
    return {
        'drift_pct': drift_pct,
        'is_perf': is_performance,
        'oos_perf': oos_performance,
        'pass': drift_pct < 2.5
    }

---

📅 Schedule & Ownership

Date	Task	Owner	Status
11-01	Data Collection	All Teams	⏳ Ready
11-03	PBO Calculation	Team PBO	⏳ Ready
11-04	DSR Calculation	Team DSR	⏳ Ready
11-05	OOS Analysis	Team OOS	⏳ Ready
11-05	Audit Verification	Team Audit	⏳ Ready
11-06	DEBT Review	Team Debt	⏳ Ready
11-07	Report Writing	All Teams	⏳ Ready
11-10	Integration & Go/No-Go	Coordinator	⏳ Ready
11-12	CTO/CFO Approval	Management	⏳ Ready
11-15	Phase 3 Prep	All Teams	⏳ Ready

---

🚨 Risks & Mitigation

Risk	Probability	Impact	Mitigation
PBO >= 20%	Low	High	모델 재조정 계획 준비
DSR <= 0.5	Low	High	전략 재평가 계획
Data corruption	Very Low	Critical	Backup & validation
Team delay	Medium	Medium	Daily standup, 병렬 추진
Approval delay	Low	Medium	사전 검토 회의

---

✅ AGENTS.md v16.0 준수

✅ 정공법: 근본 원인 분석 (각 메트릭 검증)
✅ 과유불급: 필요한 검증만 (추가 테스트 금지)
✅ 재현성: 모든 계산 문서화 + SQL 저장
✅ 이력성: 모든 결과 git에 저장
✅ 안정성: Go/No-Go 명확한 기준
✅ 현장감: 실제 Phase 1 데이터 사용
✅ 컴포넌트화: 5개 팀 독립 실행
✅ 기술부채: DEBT 최종 정리

---

Version: 1.0
Status: 🟢 READY FOR EXECUTION
Target Start: 2026-11-01
Target Completion: 2026-11-15
Go-Live: 2026-11-20 (Subject to Phase 2 passing)