> **Simulated example.** Simulated site with known ground truth (seed 4242). Not a field survey: the muon data and every auxiliary source were generated for this run.

# MuonSense — Subsurface Anomaly Screening Report (simulated example)

**Site:** Simulated randomised site (seed 4242)  
**Deployment (simulated):** Rail-guided muon detector, 14 stations  
**Exposure model:** nominal dwell per station vs cos^2(theta) reference

> Screening-level assessment. Combines passive cosmic-muon tomography with surface/landscape inspection, InSAR subsidence, and geological prior.

## Executive summary

- **3 candidate locations** assessed along the alignment.
- **2 HIGH** and **0 MEDIUM** subsurface-anomaly risk after multi-source fusion.
- Top priority: **Anomaly 1** — combined anomaly probability **82%**, corroborated by Surface, InSAR subsidence rate.
- Muography reconstruction detection significance: **3.6σ**, peak density deficit **0.74 g/cm³**.

## 1. Anomaly map (muography reconstruction)

- Reconstruction grid: [48, 32, 24] voxels, method `sart`.
- Detection significance: **3.6σ** vs the homogeneous-rock hypothesis (Pearson χ² standardised with Poisson moments; a scan consistent with rock scores about 0σ).
- Recovered density contrast localises low-density volume(s) within the scanned volume (rock ≈ 2.65 g/cm³; recovered minimum ≈ 1.91 g/cm³).
- Full 3D field exported as `reconstruction.json` (loadable in the web platform).

## 2. Priority locations — where to verify first

| Rank | Location | Combined risk | Tier | Muography | Corroborating sources | Recommended action |
|---|---|---|---|---|---|---|
| 1 | Anomaly 1 (-0.32, -0.32) | **82%** | HIGH | 100% | Surface, InSAR subsidence rate | Verify first: targeted boreholes / ground-penetrating radar at this location. |
| 2 | Decoy 1 (-0.42, +0.54) | **63%** | HIGH | 96% | Surface, Geological prior | Verify first: targeted boreholes / ground-penetrating radar at this location. |
| 3 | Decoy 2 (-1.15, -0.33) | **17%** | LOW-MEDIUM | 81% | Surface | Monitor; re-assess if surface or InSAR signals evolve. |

## 3. Risk insights — multi-source fusion breakdown

Each source contributes a signed log-likelihood ratio (LLR); positive argues for a subsurface anomaly, negative against. The posterior combines them with a base rate of 8%.

### Anomaly 1 — HIGH (82%)

| Source | Observation (score) | LLR |
|---|---|---|
| Muography (subsurface density) | 1.00 | +2.58 |
| Surface / landscape inspection | 0.82 | +0.77 |
| InSAR subsidence rate | 0.77 | +0.60 |
| Geological prior | 0.51 | +0.02 |

- Clear subsurface density deficit, corroborated by Surface, InSAR subsidence rate.

### Decoy 1 — HIGH (63%)

| Source | Observation (score) | LLR |
|---|---|---|
| Muography (subsurface density) | 0.96 | +2.37 |
| Surface / landscape inspection | 0.80 | +0.73 |
| InSAR subsidence rate | 0.41 | -0.19 |
| Geological prior | 0.56 | +0.09 |

- Clear subsurface density deficit, corroborated by Surface, Geological prior.

### Decoy 2 — LOW-MEDIUM (17%)

| Source | Observation (score) | LLR |
|---|---|---|
| Muography (subsurface density) | 0.81 | +1.63 |
| Surface / landscape inspection | 0.69 | +0.45 |
| InSAR subsidence rate | 0.20 | -0.66 |
| Geological prior | 0.14 | -0.58 |

- Clear subsurface density deficit, corroborated by Surface.

## Data sources fused (simulated inputs for this run)

- **Muography (subsurface density)** — max diagnostic strength LLR ±2.6
- **Surface / landscape inspection** — max diagnostic strength LLR ±1.2
- **InSAR subsidence rate** — max diagnostic strength LLR ±1.1
- **Geological prior** — max diagnostic strength LLR ±0.8

## Limitations

- Muography here is limited-angle (detector positions along one survey line, or one fixed viewpoint): plan position is localised well, depth is not; contrast is partially under-recovered. Confirm depth with boreholes or GPR.
- Auxiliary sources (surface, InSAR, geology) are **simulated** for this run, as is the muon data; in a real engagement they are ingested from field survey, satellite and GIS data.
- Absolute scale constants are demo-tuned, not calibrated to a specific site.