diff --git a/03_methodology.md b/03_methodology.md
new file mode 100644
index 0000000..7be8cc7
--- /dev/null
+++ b/03_methodology.md
@@ -0,0 +1,43 @@
+## III. Methodology
+
+### A. Dataset
+
+The foundation of our model rests on a curated dataset of:
+
+- **Anonymized insurance claim transcripts**  
+- **Internal emails between adjusters and claimants**  
+- **Call center logs with escalation flags**
+
+This dataset includes a balanced mixture of **confirmed fraudulent claims** and **validated legitimate cases**, used to both train and test our recursive linguistic model. Importantly, each data source is processed through a **human-AI recursive review loop**, where human analysts verify and adjust the resonance scores generated by our models — ensuring that subjectivity and nuance are preserved while expanding analytic scale.
+
+> *Every claim is not merely analyzed. It is recursively heard.*
+
+---
+
+### B. Analytical Tools
+
+To detect subtle patterns of deceptive intent, we apply an ensemble of forensic NLP methods:
+
+- **NLP-based Pattern Extraction**: Identifies clusters of linguistic anomalies across claim timelines.
+- **Sentiment Trajectory Mapping**: Tracks emotional evolution of narratives; distinguishes authentic distress from strategic affect.
+- **Syntax Entropy & Disfluency Detection**: Measures irregularities in syntactic flow, hesitation markers, and repair sequences.
+- **"Truth Collapse" Scoring via Recursive Witness Dynamics**: Quantifies the destabilization of narrative integrity under recursive interrogation.
+
+> *When truth collapses, it does not vanish — it echoes in recursion.*
+
+---
+
+### C. Classification Model
+
+From this analysis, we derive a **3-Zone Classification Model** based on *recursive coherence degradation*:
+
+- **Zone I — Unintentional Incoherence (Low Risk)**  
+  Language inconsistencies stem from stress, trauma, or low verbal fluency. These are not patterns of deception, but of chaos.
+
+- **Zone II — Adaptive Rationalization (Medium Risk)**  
+  Partial distortions. In this zone, claimants subconsciously reshape their story to protect self-image, omit responsibility, or preempt skepticism.
+
+- **Zone III — Deliberate Narrative Fabrication (High Risk)**  
+  Highly structured but recursively incoherent patterns — overjustification, shifting time references, and rehearsed empathy — mark deliberate deception.
+
+> *This model does not judge. It classifies where language begins to fracture under the weight of intention.*