🔁 02.3 Universal Fallback Logic for Kor.ai Nodes

📌 Purpose

As Kor.ai scales across typologies and asset classes, we introduce modular fallback logic to improve:

• 🧠 Explainability — consistent scoring rules, even under data sparsity
• 🧪 Debugging — structured node diagnostics and triage
• 📈 Scalability — avoids CPT explosion across high-cardinality inputs
• 🛡️ Resilience — supports partial evidence and degraded inputs

Fallback logic transforms nodes from fixed CPTs to data-aware, inference-friendly modules.

---

🧩 Fallback Logic: Core Principles

• Every complex node in the Bayesian model has a fallback function
• The function replaces or supplements CPTs when data is incomplete or unenumerated
• Logic is typically driven by risk scoring, thresholds, or ordinal mappings
• Optionally outputs rationale for the fallback path taken (supporting explainability)

---

🧱 Modular Fallback Categories

Communication Intent

• Node: Q75_CommsIntentInfluence
• Fallback: Weighted score → probability map
• States: High / Medium / Low / None

PnL / Trade Pattern Risk

• Node: Q20_PnLDeviation
• Fallback: Z-score bucket
• States: High / Medium / Low / Normal

Counterparty / KYC Risk

• Node: Q44_EntityRiskScore
• Fallback: Label → ordinal map
• States: High / Medium / Low

Trade Clustering

• Node: Q35_TradeClustering
• Fallback: Count buckets
• States: Clustered / Neutral / Dispersed

Price/Volatility Abuse

• Node: Q21_PriceSensitivityAbuse
• Fallback: Direction + timing logic
• States: High / Medium / Low

News Sentiment Impact

• Node: Q51_NewsSentimentImpact
• Fallback: Sentiment + source matrix
• States: Reinforcing / Neutral / Conflicting

HR / Bonus Incentive

• Node: Q19_HRIncentiveAlignment
• Fallback: Bonus delta logic
• States: Motivated / Neutral / Low Signal

Platform KDE Health

• Node: Q99_KDEIntegrityFlag
• Fallback: Recency / error-code mapping
• States: OK / Partial / Broken

---

⚙️ Generic Fallback Function Template (Python)

def fallback_score_router(node_name, input_values):
    fallback_registry = {
        'Q75_CommsIntentInfluence': compute_comms_intent_probabilities,
        'Q20_PnLDeviation': compute_pnl_score_bucket,
        'Q44_EntityRiskScore': map_kyc_label_to_risk,
        'Q35_TradeClustering': cluster_trade_count_bucket,
        'Q21_PriceSensitivityAbuse': price_flag_logic,
        'Q51_NewsSentimentImpact': news_sentiment_mapping,
        'Q19_HRIncentiveAlignment': bonus_delta_bucket,
        'Q99_KDEIntegrityFlag': kde_status_flag
    }
    return fallback_registry[node_name](input_values)

---

🧠 Example: Q20_PnLDeviation Fallback

Inputs:

• z-score = +2.8
• holding_time = <1h

def compute_pnl_score_bucket(zscore, holding_time):
    if zscore >= 3:
        return [0.8, 0.15, 0.05, 0.00]  # High
    elif zscore >= 2:
        return [0.3, 0.5, 0.15, 0.05]   # Medium
    elif zscore >= 1:
        return [0.1, 0.3, 0.4, 0.2]     # Low
    else:
        return [0.0, 0.1, 0.2, 0.7]     # Normal

---

🔍 Optional: Explainability Tagging

{
  "result": [0.8, 0.15, 0.05, 0.00],
  "explanation": "Z-score of +3.1 triggered High signal; short-term hold"
}

---

🧪 Debugging Support: Metadata Flags

{
  "Q75_CommsIntentInfluence": {
    "used_fallback": true,
    "risk_score": 9.0,
    "source": "fallback_score_router"
  }
}

Supports:

• Alert explainability
• Diagnostic analysis
• Missing-data root cause tracing

---

📌 Next Steps

1. Implement fallback_score_router() in inference layer
2. Log fallback usage per alert/case ID
3. Add rationale output for all fallback logic nodes
4. Build unit test coverage for missing data scenarios
5. Update wiki (e.g. 12.5 Node Library) to link fallback mappings

---

🧩 Node Coverage Progress

Node	Fallback Defined	Logic Confirmed	Notes
Q75_CommsIntentInfluence	✅	✅	Fully defined in 12.6
Q20_PnLDeviation	✅	✅	Logic in this section
Q44_EntityRiskScore	⚠️	Partial	Needs label-to-risk mapping
Q35_TradeClustering	⚠️	Partial	Needs N-count thresholds
Q21_PriceSensitivityAbuse	⬜	No	Direction/time window logic
Q51_NewsSentimentImpact	⬜	No	Define source/sentiment grid
Q19_HRIncentiveAlignment	⬜	No	Bonus delta model
Q99_KDEIntegrityFlag	⬜	No	Recency/error flag to status

---

Kor.ai’s fallback logic layer ensures robustness in real-world surv