Company Scoring System: Multi-Dimensional Intelligence

The Core Insight: V7 vs V12

The Big Idea: Learning from Inflection Points

Instead of quarterly filings triggering signals, scan 10-20 years of stock history to find inflection points (breakouts, collapses, sustained trends). Then look back 1-2 years at events leading up to that moment and learn what predicted success/failure.

Inflection Point Detection

Algorithmic Pattern Recognition:

def detect_inflection_points(ticker, lookback_years=10):
    """
    Scan price history to find moments of interest
    Returns: List of inflection events with scores
    """

    inflections = []

    for date in price_history:
        # Pattern 1: Flat → Breakout
        if (volatility_60d_before < 0.15 and
            return_90d_after > 0.30 and
            sustained_for_days > 180):
            inflections.append({
                'date': date,
                'type': 'breakout_success',
                'score': 9,  # Great outcome
                'return_6m': 35.0,
                'return_12m': 58.0
            })

        # Pattern 2: High → Collapse
        if (price > sma_200 * 1.8 and
            return_90d_after < -0.40 and
            max_drawdown > -0.50):
            inflections.append({
                'date': date,
                'type': 'collapse',
                'score': 1,  # Terrible outcome
                'return_6m': -45.0,
                'max_drawdown': -62.0
            })

        # Pattern 3: Sustained Bull Run
        if (consecutive_positive_months >= 6 and
            total_return_6m > 0.50 and
            sharpe_ratio > 2.0):
            inflections.append({
                'date': date,
                'type': 'sustained_success',
                'score': 8,  # Great outcome
                'return_12m': 75.0,
                'sharpe': 2.3
            })

        # Pattern 4: Recovery from Distress
        if (was_distressed_before and
            return_12m_after > 1.0 and
            volatility_normalized):
            inflections.append({
                'date': date,
                'type': 'turnaround_success',
                'score': 10,  # Exceptional
                'return_12m': 120.0
            })

    return inflections

Industry-Specific Catalysts

Biotech Examples:

# FDA Approval Success
if event_type == "approved_drug_trial_phase3" and return_12m > 0.80:
    score = 9  # Major success

# FDA Rejection Collapse
if event_type == "failed_drug_trial_phase3" and return_6m < -0.60:
    score = 1  # Disaster

Mining Examples:

# Resource Discovery Success
if event_type == "announced_resource_estimate_major" and return_12m > 1.5:
    score = 10  # Exceptional discovery

# Permit Rejection
if event_type == "denied_mining_permit" and return_6m < -0.50:
    score = 2  # Project dead

Multi-Dimensional Scoring Framework

{
  "company_score": 7.5,       // 0-10: Overall quality/outlook
  "risk_score": 6.8,          // 0-10: Risk level (10=safest)
  "success_probability": 0.72,// 0-1: Probability of positive outcome
  "speed_score": 5.0,         // 0-10: How fast catalysts play out
  "sustainability_score": 8.2,// 0-10: How long success lasts

  "predicted_return_6m": 18.5,  // % return estimate
  "predicted_return_12m": 32.0, // % return estimate
  "max_drawdown_estimate": -15.0,
  "sharpe_estimate": 1.8,

  "key_catalysts": [
    "approved_drug_trial_phase3",
    "raised_capital_large",
    "announced_strategic_partnership"
  ],

  "risk_factors": [
    "high_dilution_risk",
    "competitive_threats",
    "regulatory_uncertainty"
  ],

  "narrative": "Company shows strong catalyst momentum with FDA
               approval and secured financing. Partnership with
               Fortune 500 validates technology. Risk is moderate
               due to competitive landscape."
}

Scoring Rubric

Company Score (0-10):

• 10: Exceptional outcome (>100% return, sustained, low drawdown)
• 8-9: Great outcome (50-100% return, sustained)
• 6-7: Good outcome (20-50% return)
• 4-5: Moderate outcome (10-20% return)
• 2-3: Poor outcome (0-10% return or mild loss)
• 0-1: Terrible outcome (<-30% return or collapse)

Risk Score (0-10):

• 10: Very safe (low vol, high Sharpe, small drawdowns)
• 7-9: Moderate risk
• 4-6: Elevated risk
• 0-3: Very risky (high vol, large drawdowns)

Success Probability:

• Based on historical similar event patterns
• Calibrated: 0.8 probability → 80% actually succeeded
• Accounts for: industry, event type, magnitude, timing

Training Data Generation

def generate_scoring_training_data(start_year=2003, end_year=2023):
    """
    Scan all companies for 20 years to find inflection points
    """

    training_examples = []

    for company in all_public_companies:
        # Get full price history
        prices = get_price_history(company, start_year, end_year)

        # Detect inflection points
        inflections = detect_inflection_points(company, prices)

        for inflection in inflections:
            # Look back 1-2 years for events
            lookback_start = inflection['date'] - timedelta(days=365*2)
            events = get_events(
                company,
                start_date=lookback_start,
                end_date=inflection['date']
            )

            if not events:
                continue  # Skip if no events to learn from

            # Calculate outcome scores
            outcomes = calculate_outcome_scores(
                company,
                inflection['date'],
                forward_periods=[90, 180, 365]
            )

            # Create training example
            example = {
                'cik': company.cik,
                'inflection_date': inflection['date'],
                'inflection_type': inflection['type'],

                # Prompt: 1-2 year event history
                'prompt': format_events_to_prompt(events, depth='deep'),

                # Labels: Multi-dimensional scores
                'company_score': outcomes.company_score,
                'risk_score': outcomes.risk_score,
                'success_probability': outcomes.success_probability,
                'speed_score': outcomes.speed_score,
                'sustainability_score': outcomes.sustainability_score,

                # Ground truth returns
                'actual_return_6m': outcomes.return_6m,
                'actual_return_12m': outcomes.return_12m,
                'actual_sharpe': outcomes.sharpe,
                'actual_max_drawdown': outcomes.max_drawdown,

                # Event narrative
                'key_catalysts': identify_key_catalysts(events),
                'risk_factors': identify_risk_factors(events),
            }

            training_examples.append(example)

    return training_examples

Real-World Applications

Company: MRNA (Moderna)
Current Score: 8.2 / 10
Risk Score: 6.5 / 10 (Moderate)
Success Probability: 74%

Recent Events:
- approved_drug_trial_phase3 (2 weeks ago)
- raised_capital_large (1 month ago)
- announced_strategic_partnership (3 months ago)

Outlook: Strong momentum with FDA approval and financing secured.
Alert: Monitor for competitive threats in mRNA space.

Company: XYZ Corp
Risk Score: 3.2 / 10 (HIGH RISK)
Distress Probability: 45%

Warning Signals:
- covenant_violation (recent)
- material_weakness (audit)
- workforce_reduction (3 quarters)

Recommendation: Increase credit monitoring frequency

-- Find high-quality opportunities
SELECT company_name, score, risk_score, success_probability
FROM company_scores
WHERE company_score >= 7.0
  AND risk_score >= 6.0
  AND success_probability >= 0.70
  AND industry = 'Biotech'
ORDER BY company_score DESC

ALERT: 3 holdings scored below 5.0 (review needed)

AAPL: 8.5 → 7.2 (↓1.3) - Still strong
MRNA: 7.8 → 8.2 (↑0.4) - Improving
XYZ:  6.2 → 4.1 (↓2.1) ⚠️ REVIEW URGENTLY

Why This Could Work Better Than V7

Advantages:

1. More Training Data:
   • 10-20 years × thousands of companies = huge dataset
   • Not locked to quarterly filings
   • Learn from all important moments
2. Richer Context:
   • 1-2 year event history vs 180 days
   • See full catalyst buildups
   • Understand multi-quarter narratives
3. Better Labels:
   • Multi-dimensional scores vs binary profitable/not
   • Industry-specific outcomes
   • Risk-adjusted success metrics
4. Industry Specialization:
   • Train separate models: biotech, mining, tech, retail
   • Learn industry-specific catalyst patterns
   • Better accuracy per sector
5. Broader Applications:
   • Trading (like V7)
   • Credit analysis
   • Portfolio monitoring
   • Investment screening
   • Company research

Implementation Path

Phase 1: Inflection Point Detection (2-3 weeks)

• Build inflection point scanner
• Scan 10-20 years of price data
• Classify: breakouts, collapses, sustained trends, recoveries
• Validate manually on known success/failure stories

Phase 2: Training Data Generation (1 week)

• For each inflection point, gather 1-2 year event history
• Calculate multi-dimensional outcome scores
• Generate training examples (target: 15K-25K)
• Split train/val/test

Phase 3: Model Training (1 week)

• Use same nanochat architecture as V7
• Train on scoring task (not just BUY/SELL/SKIP)
• Validate score calibration
• Test on held-out companies

Phase 4: Product Integration (2 weeks)

• Build scoring API
• Monthly batch scoring for all companies
• Historical score tracking
• Alert system for score changes

Total: ~6-8 weeks after V7 proven

Success Metrics

Model Performance:

• Score prediction accuracy within ±1 point: >70%
• Success probability calibration: 0.7 → 70% actual success
• Risk score correlation with actual volatility: r > 0.6
• Better than baseline (simple momentum): +10% accuracy

Product Success:

• Raul satisfaction with company scoring dashboard
• Credit team finds it useful for monitoring
• Portfolio managers use for screening
• Customer retention if sold as product

Pricing Strategy

Standalone Product:

• Company Scoring API: $X/month (unlimited scoring)
• Historical scores included
• Monthly updates
• Alert system

Or Bundle with V7:

• V7 Signals + Company Scoring: 1.5x price
• "Complete company intelligence package"

Why Raul Cares About This