# Warren AI - Decisioni Chiave & Architettura Sistema

**Versione**: 1.0  
**Data consolidamento**: 3 Dicembre 2025  
**Fonte**: 5 verbali Nov-Dic 2025  

---

## 🎯 DECISIONI STRATEGICHE FONDAMENTALI

### 1. Ruoli Definiti (2 Dicembre 2025)

**Warren AI = Primary Engine**:
- Scan settimanale 150 titoli (IT, FR, DE, USA)
- Algoritmo deterministico proprietario ("Coca Cola recipe")
- Output: 2-5 BUY candidates (score ≥80, margin ≥20%)
- **NON fa**: News real-time, governance check, timing sentiment

**Claude = Safety Net + Strategic Advisor**:
- Pre-acquisition check: News 30gg, red flags, timing, sentiment → GO/WAIT/RED FLAG
- Market outlook strategico: Valuation levels, macro trends, scenario analysis
- **NON fa**: Stock picking indipendente, overrule Warren senza dati

**Mauro = Final Authority**:
- Decisione finale data-driven (Warren score + Claude context)
- Approvazione modifiche algoritmo
- Strategic calls (sizing, thresholds)

### 2. Filosofia Validazione Algoritmo (3 Dicembre 2024)

**70% Oggettivo (NON TOCCARE)**:
- ROE ≥15%, D/E <1.0, P/E <15x, Margin ≥20%, FCF Yield ≥4%
- Time-tested principles (80+ anni Buffett/Graham)
- Se falliscono → STOP algorithm (structural break)

**30% Soggettivo (CALIBRARE empiricamente)**:
- Country penalty (IT -20%, FR/DE -10%, USA 0%)
- Growth cap (Mature 4%, Others 5%)
- Graham multiplier per sector (Banks 15x, Luxury 30x)

**Approccio**: Observe 6-12 mesi → Validate → Calibrate se necessario → Expand se successo

### 3. Decision Point Gennaio 2026 (2 Dicembre 2025)

**Bivio Strategico**:
- **Opzione A**: Satellite 20k (R&D mode, test algorithm)
- **Opzione B**: Satellite 60-70k (active outperformance, Warren must deliver)

**Input per decisione** (Dicembre 2025):
1. Latest Warren scan + opportunity set
2. Market environment assessment (valuation levels, macro, sentiment)
3. Risk appetite Mauro aggiornato
4. Scenario probabilities 2026 (bull/base/bear)

**Trigger scale-up mid-2026**:
- Q1 correction ≥10% + warren_buy_list ≥5 → "OPPORTUNITY"
- Q2 deployed ≥80% + portfolio_up ≥8% → "SUCCESS"
- Q3 still zero buys + market_up ≥15% → "UNDERPERFORM review"

---

## 🏗️ ARCHITETTURA SISTEMA

### Schema v4.1 - Enhanced Tracking (1 Dicembre 2025)

**Database**: 78 colonne `opportunity_detailed`
- Score breakdown (6 componenti)
- Fair value methods (5 metodi + weights + adjustments)
- Algorithm parameters (growth caps, penalties, sector flags)
- Advanced quality metrics (ROIC, Interest Coverage, F-Score)

**Output Dual-Purpose**:
```
HTML (Human)     → Visual, concise, quick decisions
JSON (AI)        → Complete raw data, full breakdown, audit transparency
Database (Audit) → Time series, statistical validation
```

**JSON Structure Chiave**:
```json
{
  "score": {
    "breakdown": {
      "valuation": {"pe_score": X, "pb_score": Y, ...},
      "quality": {"roe_score": X, "debt_score": Y},
      "bonuses": {"margin_bonus": X, ...},
      "penalties": {...}
    }
  },
  "algorithm_parameters": {
    "growth_rate_used": 4.0,
    "is_growth_capped": true,
    "country_penalty_pct": 0.20,
    "graham_multiplier": 16.0
  },
  "valuation": {
    "fair_value_methods": {"pe_based": X, "pb_based": Y, ...},
    "method_weights": {"pe_weight": 0.35, ...},
    "adjustments": {
      "base_fair_value": X,
      "country_penalty": 0.20,
      "quality_premium": 0.30
    }
  },
  "advanced_quality_metrics": {
    "roic_percent": 18.5,
    "interest_coverage_ratio": 8.2,
    "piotroski_fscore": 8
  }
}
```

### Coverage Metriche (Schema v4 - 2 Dicembre 2024)

**18 Indicatori Fondamentali** (~95% coverage):

**Valuation**: P/E, P/B, P/S, Dividend Yield  
**Quality**: ROE, ROA, Current Ratio, Profit Margin  
**Growth**: Revenue Growth, Earnings Growth  
**Debt**: Debt/Equity, Net Debt/EBITDA  
**Cash Flow**: FCF, FCF Yield  
**Forward**: PEG, EV/EBITDA  
**Advanced** (v4): ROIC, Interest Coverage, Piotroski F-Score

**Coverage Raggiunta**:
- ROIC: 100% (40/40 mercato IT)
- Interest Coverage: 100% adjusted (23/23 non-Financial/Utility)
- Piotroski F-Score: 100% (40/40)

---

## ⚠️ BUG NOTI & OTTIMIZZAZIONI IDENTIFICATE

### 6 Potenziali Falle Identificate (1 Dicembre 2025)

**DECISIONE**: WAIT 8 settimane (audit 26 Gennaio 2026)  
**Rationale**: Zero ground truth, data > theory, conservative > aggressive

1. **Growth Cap Conservativo Mature Sectors**:
   - Energy/Financial: Growth capped a 4% (nominal GDP)
   - Esempio: ENI growth reale 6% ma usa 4% → fair value ridotto ~33%
   - Impatto: Score 58 potenziale 68 (border BUY)

2. **Country Penalty Italia -20%**:
   - Tutti .MI penalizzati -20% fair value
   - Export-heavy (Ferrari 90%, ENI 80%+) penalizzati come domestic
   - Esempio: Ferrari perde €40.80 FV, Azimut perde €8.03

3. **Graham Multiplier Fisso Financial (15x)**:
   - Financial: Multiplier 15x fisso (no growth premium)
   - Banche in crescita penalizzate vs mature banks
   - Esempio: UCG growth 8% valued con 15x statico

4. **AZM Classificato is_financial (BUG)**:
   - Azimut: is_financial=True → usa P/B per debt score
   - P/B 2.65 → 0/15 punti MA ha net cash €439M
   - Asset management ≠ Banca tradizionale
   - Impact: Debt score 0 invece 15, margin +1.3% potenziale +23.8%
   - **Fix posticipato**: Validare pattern generale, avoid hardcoded fix

5. **Debt Penalty Threshold 3x EBITDA**:
   - Energy/Industrial: Debt penalty >3x troppo stretto?
   - Capital intensive naturalmente leveraged
   - ENI penalizzato per debt coverage normale settore

6. **Luxury Brand Graham Cap 30x**:
   - Ferrari: P/E 37.6x vs Graham 25.8x → sempre AVOID
   - Market valuta luxury 35-40x strutturalmente
   - Algorithm troppo conservativo su luxury persistent premium?

### Mercato USA - Calibrazione Necessaria (28 Novembre 2025)

**Problema Strutturale**: Metriche europee inadatte a mercato USA moderno

**False Positives Critici**:
- Apple: CRITICAL DANGER per D/E 1.52 MA net cash $56B
- Home Depot: CRITICAL DANGER per D/E 5.45 MA ROE 163% (buyback aggressivi)
- Goldman Sachs: D/E 586x MA Tier 1 Capital 14.5% (D/E non applicabile banche)

**Differenze Strutturali USA vs EU**:
- Buyback: 50-70% utili (vs 10-20% EU) → ROE artificialmente alto
- D/E medio S&P500: 1.8x → Soglia 1.5x troppo bassa
- Tax optimization aggressiva → Debito > cassa offshore (Apple case)
- Settori dominanti: Tech capital-light vs EU industrial

**Raccomandazione**: 
- USA base D/E threshold: 2.5x (vs 1.5x EU)
- Financial Services: Use Tier 1 Capital Ratio (disable D/E)
- Tech con net cash: Ignora D/E completamente
- ROE max: 200% (vs 100% attuale) per capital-light businesses

**Status**: NON implementato, wait audit Jan 2026 per validation congiunta

---

## 📊 SCORING SYSTEM v4

### Formula Completa

```
Score Totale = min(100, (Raw Score / 130) × 100)

Raw Score Components:
├─ Valuation Score: 30 pts max (P/E 15 + P/B 10 + Div 5)
├─ Quality Score: 40 pts max (ROE 25 + Debt 15)
├─ Growth Score: 30 pts max (Revenue 15 + Earnings 15)
├─ Bonuses: +20 pts max (Margin 10, Debt Cov 5, FCF 3, PEG 1, EV/EBITDA 1)
├─ Penalties: -30 pts max (ROE negative, Debt excess)
└─ Advanced Quality: +20 pts max (ROIC 10, Int.Cov 5, F-Score 5)

Range: 0-130 raw → normalized 0-100
```

### Classification Thresholds

```
85-100: STRONG BUY (quality + safety margin eccellente)
80-84:  BUY (solido, margin adeguato)
60-79:  HOLD (quality OK, valuation cara O margin basso)
40-59:  AVOID (quality debole O overvalued)
0-39:   CRITICAL DANGER (red flags gravi)

Special: DATA INSUFFICIENT (dati mancanti critici)
```

### Metriche Avanzate (Schema v4)

**ROIC (Return on Invested Capital)**:
```
Formula: (NOPAT / Invested Capital) × 100
NOPAT = EBIT × (1 - Tax Rate)
IC = Total Assets - Current Liabilities - Cash

Thresholds:
≥15%: Eccellente (10 punti)
≥10%: Buono (7 punti)
≥5%: Accettabile (4 punti)
<5%: Debole (0 punti)

Fallback: 3 livelli (EBIT ricostruito, IC alternativo, Equity-only)
```

**Interest Coverage Ratio**:
```
Formula: EBIT / Interest Expense

Thresholds:
≥5.0x: Forte (5 punti)
≥3.0x: Adeguata (3 punti)
≥1.5x: Rischiosa (1 punto)
<1.5x: Critica (0 punti)

Sector Exclusions: Financial Services, Utilities (None = neutral 2.5 pts)
```

**Piotroski F-Score**:
```
9 Criteri Binari (1 se soddisfatto):

Profitability (4):
1. ROA > 0
2. Operating Cash Flow > 0
3. ΔROA > 0 (YoY improvement)
4. Accruals < 0 (OCF > Net Income)

Leverage (3):
5. ΔLong-term Debt ≤ 0
6. ΔCurrent Ratio > 0
7. No New Equity issued

Efficiency (2):
8. ΔGross Margin > 0
9. ΔAsset Turnover > 0

Thresholds:
8-9: Forte (5 punti)
6-7: Buono (3 punti)
4-5: Neutro (1 punto)
0-3: Debole (0 punti)

Evidence: Alpha 7.5% annuo su 20 anni backtest accademico
```

---

## 🔍 WORKFLOW OPERATIVO

### Scan Settimanale (Domenica 19:00)

```
1. Automated scan: python warren_scan.py --market ALL
2. Email report con quality alerts
3. Output generati:
   - HTML reports (human-oriented)
   - JSON reports (AI-oriented, 165KB per mercato)
   - Database insert (78 colonne tracking)
```

### Pre-Acquisition Check (Workflow con Claude)

```
Mauro: "Claude, pre-acquisition check [TICKER]"

Claude workflow:
1. Load latest JSON scan
2. Score breakdown analysis (identifica bottleneck)
3. Web search:
   - News ultime 30gg rilevanti
   - Red flags (legal, governance, downgrade)
   - Timing tecnico (overbought/oversold)
   - Sentiment analisti
4. Qualitative crash test:
   - Governance deep dive
   - Legal/regulatory risks
   - Business model sustainability
   - Competitive position

Output: GO / WAIT / RED FLAG + rationale + entry price suggestion
Tempo: 5-10 min
```

### Decision Point Strategici

**Dicembre 2025** (~4 settimane da ora):
```
Mauro: "Claude, market decision report 2026"

Claude prepara:
├─ Valuation assessment (S&P, FTSE MIB P/E vs historical)
├─ Macro trends (Fed/ECB policy, EUR/USD, inflation)
├─ Sentiment indicators (VIX, put/call, fund flows)
├─ Warren opportunity set (latest scan, avg margin, sectors)
├─ 3 Scenarios 2026:
│   ├─ Bull case (prob %, impact satellite 20k/60k)
│   ├─ Base case (prob %, impact satellite 20k/60k)
│   └─ Bear case (prob %, impact satellite 20k/60k)
└─ Raccomandazione: 20k vs 60k satellite + rationale

Tempo prep: 30-45 min
```

**Monitoring Trimestrale 2026**:
```
Q1, Q2, Q3, Q4:
├─ "Claude, satellite performance review Q[X]"
├─ Output:
│   ├─ Performance vs ETF benchmark
│   ├─ Warren hit rate (% BUY che funzionano)
│   ├─ Market conditions update
│   └─ Raccomandazione: Stay 20k / Scale 60k / Reduce / Abort
└─ Trigger scale-up (se correction materializes + warren buys)
```

### Audit Algoritmo (26 Gennaio 2026 - Week 8)

**Checkpoint Critico**: Validazione 6 potenziali falle con 8 settimane dati

**Query SQL da eseguire** (`audit_queries.sql`):
```sql
-- 1. Quality Check: High scorers = high ROE?
-- 2. Country Penalty Check: Italy avg_margin vs score
-- 3. False Positive Rate: BUY recommendations performance
-- 4. Sector Bias: Settori ignorati sistematicamente?
-- 5. Growth Cap Impact: Missed winners per growth capping?
-- ... (10 query totali)
```

**Decision Tree**:
```
IF false_positive_rate < 15%:
    ✅ Algoritmo OK, parametri conservativi CORRETTI
    → NO fix needed, keep v4.1
    
ELIF false_positive_rate 15-30%:
    ⚠️ Algoritmo troppo permissivo in alcune aree
    → Deploy v4.2 con fix validati empiricamente
    
ELSE (>30%):
    🔴 Algoritmo broken, fix NECESSARI
    → Deep review + v4.2 comprehensive fix
```

---

## 💡 KEY LEARNINGS & BEST PRACTICES

### 1. Architecture: Data FIRST, Classification AFTER

**Problema identificato**: 15-40% titoli non calcolavano metriche avanzate per early exit

**Soluzione**:
```python
# ❌ SBAGLIATO
if is_critical:
    return {'rating': 'CRITICAL', 'metrics': None}
calculate_metrics()  # Mai raggiunto

# ✅ CORRETTO
metrics = calculate_all_metrics()  # SEMPRE eseguito
rating = classify(metrics)
return {'rating': rating, 'metrics': metrics}
```

**Principio**: Calculate data FIRST, determine classification AFTER

### 2. Fallback Graceful > Rigid Logic

**Esempio ROIC**:
1. Metodo standard (EBIT diretto)
2. Fallback Livello 1 (EBIT ricostruito: Net Income + Interest + Tax)
3. Fallback Livello 2 (Invested Capital alternativo: Debt + Equity - Cash)
4. Ultra-fallback (solo Equity)

**Risultato**: Coverage ROIC 62% → 100%

### 3. Sector-Aware Logic > Universal Thresholds

**Esempio Interest Coverage**:
- Industrial/Tech: EBIT / Interest Expense (standard)
- Financial Services: None (usa Tier 1 Capital Ratio)
- Utilities: None se configurato (regulated OK)

**None ≠ sempre "dato mancante"**, a volte = "non applicabile"

### 4. Conservative Bias = Feature, Not Bug

**Esempio**:
- Ferrari: ROIC 31%, margins 42%, moat 10/10 MA P/E 37x → AVOID
- BC.MI: ROIC 32%, Int.Cov 12.8x MA D/E 2.51 → CRITICAL DANGER

**Verdetto Warren AI**: CORRETTO. Quality eccellente + Valuation folle = AVOID sempre.

**Warren Buffett**: "It's far better to buy a wonderful company at a fair price than a fair company at a wonderful price."

**Sistema preferisce**: False negatives (missed opportunities) > False positives (losing money)

### 5. Empirical Validation > Theoretical Optimization

**6 falle identificate** (growth cap, country penalty, Graham multiplier, etc.)

**Decision**: WAIT 8 settimane (1200 data points) invece di fix immediati

**Rationale**:
- Zero ground truth su cosa funziona realmente
- Risk overfitting su teoria senza dati
- Conservative approach meglio di aggressive optimization prematura

**Quote**: "Algorithm optimization is a device for transferring money from the overfitters to the data-driven."

---

## 📈 CASI STUDIO RILEVANTI

### Caso 1: Azimut (AZM.MI) - Score 81 ma HOLD

**Analisi JSON**:
```
Score: 81/100 (raw 91 normalizzato)
├─ Valuation: 24/30 ✅ (P/E 15, P/B 4, Div 5)
├─ Quality: 25/40 ⚠️ (ROE 25, Debt 0!)  ← Bug
├─ Growth: 27/30 ✅
├─ Bonuses: +15 ✅
└─ Penalties: 0 ✅
```

**Bottleneck identificati**:
1. **Margin Too Low**: +1.3% (serve ≥20% per BUY)
2. **Country Penalty**: -20% toglie €8.03 FV → se -10%: margin +13.9%
3. **Bug is_financial**: Debt score 0/15 invece 15/15 (net cash €439M)

**Simulazione fix**:
```
AZM con fix:
├─ is_financial → False (usa D/E non P/B)
├─ Debt score: 0 → 15 pts
├─ Country penalty: -20% → -10%
├─ Score: 81 → ~96
├─ Fair value: €35.35 → €43.20 (+22%)
└─ Margin: +1.3% → +23.8% → ✅ STRONG BUY
```

**Holding attuale**: 176 azioni @ €34.26 = €6,029 (30% Satellite)

**Thesis**: ROE 29%, Net Cash €439M, Margins 78%/42%, Asset Management quality

**Raccomandazione**: HOLD position, accumulo disabled fino 1/1/2026

### Caso 2: Ferrari (RACE.MI) - Quality 10/10 ma AVOID

**Score**: 61/100 (HOLD border AVOID)

**Fair Value Breakdown**:
```
5 Valuation Methods (Weighted):
├─ P/E Based: €229.88 (peso 35%) = €80.46
├─ P/B Based: €126.14 (peso 25%) = €31.54
├─ FCF Yield: €100.44 (peso 20%) = €20.09
├─ EV/EBITDA: €256.21 (peso 15%) = €38.43
└─ Dividend: €669.60 (peso 5%) = €33.48
   = Weighted Avg: €203.99

Adjustments:
├─ Base FV: €203.99
├─ + Quality Premium: +30% = €265.19
└─ - Country Penalty: -20% = €212.15

Current: €334.80
Margin: -57.8% 🔴 OVERVALUED
```

**Algorithm Parameters**:
- Graham Multiplier: 25.8x (luxury premium included)
- P/E Actual: 37.6x (trades at 1.5x fair P/E)
- is_luxury: TRUE ✅

**Verdetto**:
```
┌──────────────────────────────────────────────┐
│ FERRARI: AVOID at €335                       │
│                                              │
│ Quality: 10/10 ⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐               │
│ Valuation: 2/10 🔴                           │
│                                              │
│ "Wonderful company, WRONG price"             │
│                                              │
│ Entry Points:                                │
│   Conservative (20% margin): €170 (-49%)     │
│   Fair value: €212 (-37%)                    │
└──────────────────────────────────────────────┘
```

**Warren AI corretto**: Quality eccellente MA P/E 37x folle per value investing.

### Caso 3: BC.MI (Brunello Cucinelli) - CRITICAL MA ROIC 32%

**Prima del fix Schema v4**:
```
Rating: CRITICAL DANGER (D/E 2.51 > 1.5)
Metriche: ROIC=None, Int.Cov=None, F-Score=None
```

**Dopo fix Schema v4**:
```
Rating: CRITICAL DANGER (D/E still 2.51)
Metriche:
├─ ROIC: 31.89% 🔥 Top 5% efficienza capitale
├─ Interest Coverage: 12.82x 🔥 Può pagare interessi 12 volte
└─ F-Score: 5/9 (quality media-buona)
```

**Insight Critico**:
- **Classification**: CRITICAL per D/E alto
- **Reality**: Business quality eccellente, leverage alto MA sostenibile
- **Distinzione**: "Leverage alto ma gestibile" ≠ "Azienda in distress"

**Learning**: Metriche avanzate permettono di distinguere situazioni che appaiono simili ma sono fondamentalmente diverse.

---

## 🚨 CONSTRAINT OPERATIVI ATTUALI

### Freeze Acquisti (fino 1 Gennaio 2026)

- **DIVIETO ASSOLUTO**: Zero acquisti fino 1/1/2026
- **Eccezione vendita**: Solo thesis breaker Azimut (ROE <10%, scandal, management change)
- **Modalità**: Observer Mode (monitor, analyze, prepare, NO execute)
- **Rationale**: R&D phase - validare algoritmo senza rumore decisionale

### Watchlist Q1 2026

**Status**: Da definire Dicembre 2025 con scan settimanali

**Criteri selezione**:
- Score ≥75
- Margin ≥10%
- Diversificazione settoriale vs Azimut (no doppio financial services)
- Budget max €3.5k per titolo

**Action permessa**: Qualify candidates (fundamentals, governance, legal DD)  
**Action vietata**: Execute trades, trim Azimut (salvo thesis breaker)

### Azimut Position Management

- **Holding**: 176 azioni @ €34.26 = €6,029 (30% Satellite)
- **Thesis**: ROE 29%, Net Cash €439M, Margins 78%/42%
- **Trim trigger**: Solo se >€38 AND margin <-15% AND conferma esplicita
- **Accumulo**: Disabled fino 1/1/2026, poi valuta se <€32 (margin +6%)
- **Monitoring**: Earnings Q4 ~15 Dicembre - verify ROE, guidance 2026

---

## 📅 TIMELINE 2025-2027

### Dicembre 2025 (CORRENTE)
- ~15 Dic: 🔴 Azimut Q4 earnings (thesis validation)
- 29 Dic: Final watchlist Q1 2026
- 31 Dic: Market decision report 2026

### Gennaio 2026
- 1 Gen: 🔓 FREEZE END (acquisti riabilitati)
- 5 Gen: Primo scan post-freeze + decision deployment
- 26 Gen: 🔴 AUDIT Week 8 algoritmo (validation 6 falle)

### 2026 (Quarterly Reviews)
- 31 Mar: Q1 Review (performance, hit rate, trigger scale-up?)
- 30 Giu: Q2 Review (scale-up evaluation)
- 30 Set: Q3 Review
- 31 Dic: 🔴 ANNUAL REVIEW + Decision satellite 2027

### Gennaio 2027
- **Decision finale**: Satellite 20k vs 60-70k (basato su performance 2026)

---

## 📚 FILES REFERENCE

### Dati Scan (PRIORITÀ ALTA per Claude)
- `reports/latest/json/warren_scan_IT_latest.json` (165KB, 40 stocks)
- `reports/latest/json/warren_scan_FR_latest.json`
- `reports/latest/json/warren_scan_DE_latest.json`
- `reports/latest/json/warren_scan_USA_latest.json`

### Core Architecture
- `DIARIO_DI_BORDO.md`: Storia completa, troubleshooting, evolution
- `SCORING_ENGINE.md`: Formula proprietaria completa
- `PROJECT_INSTRUCTIONS.md`: Bootstrap instructions Claude
- `audit_queries.sql`: 10 query validation algoritmo

### Database
- `data/trading_system.db`: Schema v4.1 (78 colonne)
- Backup: `data/trading_system.db.backup_v4.1_20251201_1830`

---

## 🎯 SUCCESS METRICS (KPI)

### Satellite Performance (2025-2026)

**Anno 1 = R&D Phase** (non performance chase):

| Metrica | Target | Rationale |
|---------|--------|-----------|
| **Capital preservation** | ≥0% return | Priority #1 in R&D |
| **Quality accuracy** | Top 20% scores → Top 30% ROE | Algorithm identifica quality |
| **Thesis breaker rate** | ≤20% holdings | Pochi errori gravi |
| **Concentration** | ≤40% single stock, ≤60% sector | Risk management |
| **Avg holding period** | ≥12 mesi | Long-term, non trading |

**NON misuriamo** (troppo presto): Sharpe ratio vs S&P, Alpha/Beta, Win rate %

### Algorithm Validation (Week 8 - Jan 2026)

| Metrica | Target | Action |
|---------|--------|--------|
| **False Positive Rate BUY** | <15% | ✅ Keep v4.1 |
| **False Positive Rate BUY** | 15-30% | ⚠️ Deploy v4.2 fix |
| **False Positive Rate BUY** | >30% | 🔴 Deep review |
| **Average Return BUY** | Positivo | ✅ Algoritmo working |
| **BUY Count 8 weeks** | 5-10 | ✅ Conservative OK |
| **BUY Count 8 weeks** | 0-2 | 🔴 Troppo conservativo |

---

## 💬 CITAZIONI CHIAVE

### Warren Buffett

> "Rule No. 1: Never lose money. Rule No. 2: Never forget rule No. 1."

> "It's far better to buy a wonderful company at a fair price than a fair company at a wonderful price."

> "The stock market is a device for transferring money from the impatient to the patient."

> "Only when the tide goes out do you discover who's been swimming naked."

> "Price is what you pay. Value is what you get."

### Filosofia Warren AI

> "Algorithm optimization is a device for transferring money from the overfitters to the data-driven."

> "Measure twice, cut once. 8 settimane di dati valgono 100 ipotesi teoriche."

> "Claude non può tirarti fuori i titoli dal cappello. Quello rimane il compito di Warren: Claude però ti può aiutare in fase di acquisto a fare check preliminare." - Mauro

> "80% dei casi Warren basta. 20% dei casi Claude utile (dubbi, timing, eventi esterni)." - Accordo workflow

> "Difendiamo i principles (ROE, FCF, P/E). Calibriamo le assumptions (country risk, growth cap). Se i principles falliscono → stop. Se le assumptions falliscono → adjust."

---

**Fine Documento Consolidato**  
**Versione**: 1.0  
**Schema Warren AI**: v4.1 (Enhanced Tracking)  
**Last Updated**: 3 Dicembre 2025

*This document consolidates 5 verbali (Nov-Dic 2025) into essential decisions and knowledge.*