🗣️ FedSpeak — Policy Signals → Structured Intelligence

---

---

📘 Overview

FedSpeak is a governed, CPMAI-aligned Policy Language Intelligence Engine that transforms
Federal Reserve communication—Beige Book, FOMC Statements, Minutes—into structured signals suitable for:

• Macro fusion systems (e.g., the_Spine)
• RegTech, FinTech, and supervisory analytics
• Policy-drift and uncertainty diagnostics
• Scenario-aware macro narrative engines

FedSpeak canonically processes Federal Reserve text and outputs:

• Policy-risk features: inflation_pressure, growth_concern, policy_uncertainty, policy_bias
• Stability diagnostics: rolling uncertainty, coherence drift
• Regime probabilities: Disinflation Recovery, Soft-Landing Drift, Stagflation Variant, Illusory Wealth Regime

It produces explainable, reproducible, research-grade features backed by strict metadata, schema enforcement, and deterministic computation models.

---

⭐ Key Features

🔹 Policy Signal Extraction

• Inflation, growth, uncertainty, and coherence metrics
• Hawkish/dovish classification
• Structured speech + statement parsing

🔹 Beige Book NLP Engine

• Canonical sentence extraction
• LDA + RBL topic modeling
• Tone classification across business, labor, wages, and prices

🔹 Minutes & Statement Processing

• Risk balance scoring
• Tone-weighted paragraph inference
• Policy-bias quantification

🔹 Drift & Stability Metrics

• Uncertainty drift
• Coherence drift
• Rolling stability windows

🔹 Regime Scoring Engine

Uses fused signals to generate regime probabilities:

• Disinflation_Recovery
• Soft_Landing_Risk_Drift
• Stagflation_Variant
• Illusory_Wealth_Regime
• Unknown

🔹 Fully CPMAI-Aligned

• Deterministic logic
• Audit-ready metadata
• Schema version enforcement
• No external APIs required

---

🧠 the_Spine • Return

---

📦 Repository Structure

fed_speak/
├── data_ingest/                   # Download & ingest Fed texts
├── preprocess/                    # Canonicalization and segmentation
├── sentiment/                     # Fed-adjusted sentiment scoring
├── features/                      # Beige/Minutes/Statement feature builders
├── drift/                         # Drift + stability metrics
├── fusion_fed_speak.py            # Composite features + regime engine
├── scripts/
│   ├── run_minutes_to_signals.py
│   ├── run_statements_features_for_spine.py
│   ├── run_beige_features_for_spine.py
│   ├── build_fed_only_macro_state.py
│   └── build_fed_drift_metrics.py
└── utils/
    ├── metadata.py
    └── schema_checks.py


# 🧠 Key Outputs

| File | Description |
|------|-------------|
| `canonical_sentences.parquet` | Cleaned, structured policy-relevant units |
| `sentiment_scores.parquet` | Fed-adjusted sentiment values |
| `beige_topics.parquet` | LDA topic distributions |
| `beige_topics_rbl.parquet` | Top-slice RBL (interpretable topics) |
| `combined_policy_leaf.parquet` | Final policy-risk leaf (∈ [-1, 1]) |

These artifacts support:
- Macro-state fusion  
- Policy-drift narratives  
- Risk diagnostics  
- Scenario commentary  
- Cross-asset interpretation  

---

# ⚙️ Usage

### Run the full Beige Book → Policy Leaf pipeline
```bash
python -m fed_speak.run_tranche1_pipeline