How MineGuard Works
A complete technical walkthrough of the three-layer verification system — from baseline establishment to statistical theft detection.
1. Baseline Establishment
Every MineGuard deployment starts with a supervised baseline period of 5-7 days. During this time, the operation owner (or a trusted representative) is physically present to ensure all measurements are honest. The baseline captures both ground richness (via daily scoop assays) and equipment performance (runtime, fuel rates, recovery rates) — both are needed to set meaningful detection thresholds.
What Gets Recorded
- Ground richness via scoop assays (g/t)
- Gold recovered at cleanout (grams per day)
- Fuel consumed (liters per day)
- Equipment run hours (per machine)
- Material processed (tons per day)
Why It Matters
The baseline establishes two critical numbers: the ground richness (ore grade in g/t from daily scoop assays) and the recovery rate for your specific equipment (typically 60-75% for sluice boxes). Together these set the expected gold output per ton processed.
Without knowing the ground richness, you can't calculate expected production — and without expected production, you can't detect skimming. The baseline must cover different material types and conditions to account for natural variation across your claim.
2. Audio Monitoring Pipeline
A weatherproof microphone mounted near each piece of equipment feeds audio to a Raspberry Pi running a machine learning classification model. The system detects three states: engine running, engine idle, and engine off.
Why Audio?
Audio monitoring is exceptionally hard to tamper with. A worker cannot fake engine sounds, and muffling the microphone is itself detectable (sudden signal drop). The ML model uses spectral features (MFCC) fed into a simple CNN classifier, achieving greater than 95% accuracy for on/off classification after site-specific calibration.
Microphone
Captures engine sounds 24/7
Raspberry Pi
On-device ML inference
MFCC + CNN
Spectral classification model
The system calculates actual runtime hours per day for each piece of equipment. These hours feed directly into the fuel verification and gold production calculations.
3. Fuel Calibration (Run-to-Empty)
Each piece of equipment is calibrated by running until fuel runs out, establishing the exact consumption rate in liters per hour at working load.
Typical consumption rates:
Generator ~4 L/hr
Excavator ~3 L/hr
Wash plant pump ~2 L/hr
Once calibrated, the expected daily fuel consumption is simply the sum of each equipment's runtime multiplied by its consumption rate:
Comparing fuel purchased against fuel expected gives the first verification layer. A large discrepancy means fuel is being siphoned off or equipment is not actually running as long as reported.
4. Daily Assay Compositing
Throughout the day, the worker scoops a representative sample of concentrate from each wash cycle. A load cell on the sampling scoop ensures consistent sample weight every time.
Compositing
Individual samples are composited (mixed together) into a daily container. This composite represents the full day's production, smoothing out natural variation between individual washes.
On-Site Assay
At the end of the day, the composite sample has been dissolving in the bleach+HCl solution for hours. The worker pours off the clear liquid, dilutes it, inserts an SPE strip, and presses a button — the electrochemical measurement takes seconds. The result is the gold concentration of the day's raw ground in grams per ton. No external lab needed.
See the Chemistry page for full details on the bleach+HCl and iodine+KI extraction methods.
5. Verification Math
The core formulas that drive MineGuard's detection engine:
Expected production
Expected Gold (g) = Material Processed (t) × Baseline Grade (g/t) × Recovery Rate
Deviation calculation
Deviation (%) = (Expected - Reported) / Expected × 100
Alert thresholds
A single day's deviation can be caused by natural factors — different material quality, equipment issues, weather. But sustained patterns over many days are statistically significant and almost certainly indicate human intervention.
6. Cross-Correlation Defense
This is the key insight that makes MineGuard powerful: all three data streams — gold grade, equipment runtime, and fuel consumption — must be consistent for a “normal” reading. A thief would need to fake all three simultaneously, which is practically impossible.
| Gold Grade | Runtime | Fuel | Diagnosis |
|---|---|---|---|
| Down | Normal | Normal | Gold skimming (human intervention) |
| Down | Down | Down | Equipment problem (legitimate) |
| Normal | Normal | Up | Fuel theft (siphoning) |
| Normal | Normal | Normal | Normal operations |
The cross-correlation means theft requires simultaneous manipulation of independent physical systems — audio signatures, fuel flow meters, and chemical assay results. This is the fundamental reason MineGuard works: the attack surface is too wide.
7. Statistical Detection Power
Individual daily measurements have noise — up to ±15% natural variance is expected. But over multiple days, random noise cancels out while systematic theft accumulates.
Standard error reduction
Standard Error = σ / √n
After 5 days: SE ≈ 6.7% of σ
After 10 days: SE ≈ 4.7% of σ
After 20 days: SE ≈ 3.4% of σ
Cumulative z-score test
z = mean_deviation / (σ / √n)
When |z| > 1.96, the system flags with 95% confidence (p < 0.05).
7-10
days to detect 20% skimming
95%
statistical confidence threshold
1/√n
error reduction per day
Think of it like flipping a biased coin. On any single flip, you cannot tell if it is biased. But after enough flips, the bias becomes obvious. MineGuard works the same way — a thief skimming 20% daily will be caught within two weeks with near-certainty.
8. Simple Daily Protocol
The system is designed so any worker can follow the daily routine without specialized training. Eight simple steps, guided by beeps and lights.
Start MineGuard System
Turn on the MineGuard system at the start of shift. Green light confirms it is running and all sensors are active.
Equipment Startup
System auto-detects engines starting via audio monitoring. No manual logging needed — runtime tracking begins automatically.
Take Scoop Samples
During wash, scoop ~20g of raw ground into the composite jar every 30 minutes. Load cell beeps when correct weight is reached. Gold dissolves passively in the bleach+HCl solution throughout the day.
Record Fuel
When adding fuel, flow sensor auto-captures the amount. No manual entry required.
Final Sample
At end of day, add the final scoop to the composite container. This represents the full day's production.
Run Assay
The composite sample has been dissolving all day. Pour off the clear liquid, dilute, insert the SPE strip and press the button. System beeps when the measurement is done.
Enter Results
System calculates and displays the day's summary — gold grade, fuel usage, runtime hours, and deviation from baseline.
Check Status
Green light means all good. Yellow means review needed. Red means an alert has been sent to the operation owner.
Dive Deeper
Explore the chemistry behind the on-site assay or see the hardware that makes it all work.