diff --git a/intel/radar/3D energy compression debug.md b/intel/radar/3D energy compression debug.md
new file mode 100644
index 0000000..2978c3a
--- /dev/null
+++ b/intel/radar/3D energy compression debug.md	
@@ -0,0 +1,77 @@
+# 🚩 Benchmark: 3D Energy Compression & Environmental Clutter (Tree Density)
+
+**Metadata:**
+- **Status:** Research & Mitigation Planning (Iteration 14)
+- **Domain:** Physics-Based Radar Simulation / ADAS Fidelity
+- **Goal:** Resolve signal intensity imbalance between thin targets (Vehicles) and voluminous targets (Trees/Buildings).
+
+---
+
+## 🔍 1. Problem Statement: The "Super-Reflector" Tree
+
+In Iteration 13, we identified a significant intensity discrepancy in the Foxglove 3D panel. While NPC vehicles (primary ADAS targets) produced stable but moderate point clouds, environmental objects like **Trees** and **Distant Buildings** generated extremely dense, high-SNR reflections that overwhelmed the CFAR detector.
+
+### Observed Discrepancies:
+- **Visual Clumping:** Trees appeared as solid "walls" of radar points, whereas in real-world 77GHz hardware, trees should appear as sparse, flickering points due to low-RCS foliage.
+- **Dynamic Range Mismatch:** The "Magnitude" attribute for a tree (voluminous) was an order of magnitude higher than a sedan (thin metallic surface), violating expected ADAS SNR profiles.
+
+---
+
+## 🧮 2. Root Cause Analysis
+
+The issue is a side-effect of how the Shenron pipeline compresses a 3D LiDAR scene into a 2D Radar heatmap. 
+
+### A. The "Shrinking" Math (`heatmap_gen_fast.py`)
+The physics engine iterates through all 100k+ LiDAR points and **sums their complex phases** into the ADC signal cube based strictly on **Range (rho)** and **Azimuth (theta)**.
+
+### B. The Vertical Summation Problem (`Sceneset.py`)
+In the current implementation (Line 412):
+```python
+P_incident = np.power(rho,2) * ... * (1/np.power(rho,tx_dist_loss_exponent)) * K_sq 
+```
+- **Infinite Vertical Gain:** The radar currently has **no vertical beamwidth** ($G(\phi)$). It "sees" every point from the ground to the sky with equal gain.
+- **Energy Integration:** A tree consists of thousands of points spread vertically. When the 3D scene is flattened, the radar treats those 1,000 vertical points as a single, super-powered reflection. 
+- **Result:** The 2D energy bin for a tree is the **sum of all its vertical slices**, making it look artificially "loud" compared to a 1.5m tall car.
+
+---
+
+## 🛠️ 3. Proposed Mitigation Strategies
+
+### Phase A: Vertical Antenna Gain (Boresight Weighting)
+Implement a **Gaussian Vertical Beam Pattern** in the `P_incident` calculation to model the directive nature of ADAS antenna arrays.
+- **Math:** Introduce a damping factor $G_{elev} = \exp\left(-2.77 \cdot \left(\frac{\phi}{\beta}\right)^2\right)$, where $\phi$ is the elevation angle and $\beta$ is the vertical beamwidth.
+- **The "2.77" Constant:** This constant ensures that when the elevation ($\phi$) reaches exactly half the beamwidth ($\beta/2$), the power is attenuated by exactly **3dB (50%)**. This aligns the simulation with standard hardware datasheets.
+- **Impact:** Points at the top of a tree or skyscraper are "blinded" by the antenna pattern, ensuring their vertical energy isn't integrated into the horizontal detection bin.
+
+### Phase B: Density-Based Normalization (The "Resolution Trap")
+Solve the physical inconsistency where increasing LiDAR resolution increases Radar intensity.
+- **The Concept:** Each LiDAR point is a "proxy" for a small piece of surface area. If the LiDAR resolution doubles (from 32 to 64 channels), each point should logically represent half the area.
+- **The Fix:** Normalize the total integrated power by the local point density: $P_{total} = \frac{1}{N} \sum P_i$.
+- **Impact:** This makes the simulation **"Sensor Independent."** A car will return the same Radar SNR whether it is sampled with 10 LiDAR points or 1,000.
+
+---
+
+## 🧭 4. Implementation Steps (Iteration 14)
+
+1. **Hardware Profile Update:** Define `self.vertical_beamwidth` in `ConfigureRadar.py`.
+2. **Physics Engine Implementation:** Add the Gaussian vertical damping logic and $1/N$ normalization to `shenron/Sceneset.py`.
+3. **DSP Calibration:** Fine-tune the `threshold` and `guard cells` in `sim_radar_utils/config.yaml` to match the new dynamic range.
+
+---
+
+## 📚 Annex: Technical Reference
+
+### Key Files to Reference
+| File | Role | Change Required |
+| :--- | :--- | :--- |
+| **`ConfigureRadar.py`** | Hardware Profiles | Define `vertical_beamwidth` for `awrl1432` and `radarbook`. |
+| **`shenron/Sceneset.py`** | Physics Engine | Implement Gaussian $G_{elev}$ and density normalization. |
+| **`sim_radar_utils/config.yaml`** | DSP Config | Adjust `threshold` and `guard cells` after vertical energy is balanced. |
+
+### Knowledge Base
+- **Calibration History:** [Shenron_debug.md](file:///d:/CARLA/CARLA_0.9.16/PythonAPI/Fox/intel/radar/Shenron_debug.md) (Iterations 01-13).
+- **Architecture Guide:** [context.md](file:///d:/CARLA/CARLA_0.9.16/PythonAPI/Fox/intel/internal/context.md) (AI navigation rules).
+
+---
+
+*Generated by Antigravity AI | 2026-04-07 | Knowledge Transfer Benchmark V2.*