# [Plan] Shenron Model-Based Development (MBD) Integration

This plan shifts the integration to a **Model-Based Development** architecture. We will treat the isolated Shenron logic as a "strictly defined black box" (similar to a Simulink model) and enhance it to output "Rich" radar data.

## User Review Required

> [!IMPORTANT]
> **Rich Point Cloud**: We will modify the extraction logic to output more than just geometry. The new format will be: `[x, y, z, velocity, SNR, RCS_estimate]`.
> 
> **Source of Data**: 
> - **SNR**: Extracted from the peak magnitude in the Range-Doppler heatmaps.
> - **RCS**: Back-calculated from the return power logic in `Sceneset`.
> - **Noise**: Sampled from the simulated noise floor in `ConfigureRadar`.

---

## Phase 1: Environment & Dependency Setup
... (Previously defined)

---

## Phase 2: The "Black Box" Wrapper & Rich Extraction

### [MODIFY] `ISOLATE/sim_radar_utils/radar_processor.py`
Modify `convert_to_pcd()` to preserve the magnitude of the detected peaks.
- **Current**: Returns `[x, y]`.
- **New**: Returns `[x, y, z, velocity, magnitude]`.

### [NEW] `ISOLATE/model_wrapper.py`
The wrapper will now handle the "Rich" data synthesis:
1. Call `Sceneset` to get the ground-truth physical power (RCS).
2. Call `HeatmapGen` to generate the noisy ADC stream.
3. Call `RadarProcessor` to extract detected peaks.
4. Merge the detected peaks with the underlying RCS data to provide a complete "Semantic Radar" point cloud.

---

## Phase 3: Modular Pipeline Integration
... (Previously defined)

---

## Verification Plan

### Automated Verification
1. **API Integrity**: Run a script that initializes the `ShenronRadarModel` and passes a dummy [N, 7] array to ensure it executes and returns the correct [N, 4] shape.
2. **Batch Run**: Execute `generate_shenron.py` on an existing dataset.

### Manual Verification
1. Compare the MCAP output with vs. without the Shenron layer to ensure topic alignment.

---

## Verification Plan

### Automated Verification
1. **Pilot Run**: Record a 100-frame scenario with semantic LiDAR.
2. **Post-Process**: Run `python scripts/generate_shenron.py`.
3. **Check Output**: Verify `data/<session>/shenron_radar/` contains 100 `.npy` files.

### Manual Verification
1. Run `python scripts/data_to_mcap.py`.
2. Open the MCAP in Foxglove and verify that the `/radar/shenron` topic perfectly overlays with the `/lidar` data but shows radar-typical noise and sparsity.