PROJECT MAGNET
Project Magnet

An instrument for magnetic anomalies.

PROJECT MAGNET turns an iPhone into a field instrument. It learns the magnetic character of wherever you're standing, watches for departures from it, and records what it measured — the shape, the strength, the place, and the moment.

Observe · Verify · Understand
MAGNETIC FIELD 165.4 +0.4 µT FROM BASELINE NORMAL
What it does

Most of the magnetic field around you is constant. The interesting part is what changes.

A magnetometer reading taken indoors is rarely the textbook Earth field. It carries the building, the wiring, the device's own magnets — an offset that can put the number three times higher than a physics lesson would suggest. Declaring "anything above X is an anomaly" is therefore useless.

So PROJECT MAGNET measures the field where you are, treats that as normal, and reports departure from it. That one decision is what makes the instrument work in a basement, a field, or a parking lot without being retuned.

10 HzMagnetometer sampling
~6 sStartup calibration
~4 sPre-trigger capture
0.5 sDwell before an event confirms
The signal path

From raw sensor to recorded observation.

Seven stages, each doing one job. The design is deliberately conservative: every stage is inspectable, and nothing is discarded that a future algorithm might want.

01SampleRaw magnetometer, ten times a second.
02MagnitudeThree axes reduced to one number, so turning the phone can't hide an anomaly.
03SmoothA moving average damps sensor jitter without blunting a real approach.
04CalibrateSix seconds of averaging establishes what "normal" means here.
05CompareDeparture from baseline, with slow drift correction that pauses during events.
06ConfirmA departure must hold for half a second. Single-sample spikes are rejected.
07RecordThe event stays open until it ends — peak, duration, shape, place.
The instrument

Built to be used while walking, not while reading.

Departure gauge

The dial shows distance from normal, with the trigger threshold marked at its midpoint. Green left, red right — readable at arm's length, in sunlight, in motion.

Geiger mode

Haptic pulses that quicken as you close on a target. The instrument can be read through your hand, so your eyes stay on the ground.

Signature capture

Every event stores the seconds before, during and after it — including the approach. An observation is a shape you can re-examine, not a number you have to trust.

Confidence engine

Amplitude against the local noise floor, persistence, consistency, prominence. Every component is shown, because a score you can't interrogate is decoration.

Survey map

Observations plotted as graduated discs — area by strength, colour by confidence. Recurring sites and strong finds are visible before you tap anything.

Diagnostics

Every internal value, live: raw, smoothed, baseline, delta, noise floor, dwell counter, detector state. The instrument you need to tune the instrument.

Principles

The credibility of the instrument is the product.

Evidence before conclusions

A large reading establishes that an unusual magnetic field was present. It does not establish what produced it, and the app never pretends otherwise. Events are classified by the shape of the signal — spike, sustained, oscillation — never by a guess at their cause.

Privacy by design

Your precise locations are yours. They stay on your device. Anything contributed to a shared network in future will carry approximate position and an anonymous identifier — never your identity, never your exact movements.

Don't destroy tomorrow's evidence

Today's detection algorithm is not the last one. Raw measurements are preserved alongside processed ones, and every event keeps its waveform, so observations recorded this year can be re-analysed by a better method later.

Confidence through corroboration

One phone reporting something unusual is a curiosity — it could be a case magnet, a passing vehicle, a speaker. Several independent devices reporting it in the same place at the same time is evidence. That distinction is the whole design of what comes next.

Straight answers

What this instrument does not yet know.

An instrument that overstates itself is worse than no instrument. Here is the honest state of the work.

  • The tuning constants are not calibrated. The thresholds, the dwell requirement, the smoothing factor and the drift rate are first-draft engineering values. They have been shown to work; they have not been optimised by controlled testing.
  • There is no published detection range. Response to a known magnet is strong and repeatable, but distance-versus-field has not been characterised, so no honest claim about range is available.
  • There is no measured false-alarm rate. Behaviour near vehicles, buildings and power lines has not been systematically surveyed.
  • Motion is flagged, not solved. Readings taken while the device was being swung are marked as suspect, because the measurement still happened — but distinguishing a genuine field change from a movement artefact remains open work.
  • Calibration can be fooled. Launch the app beside a strong source and it will learn that source as normal. Manual recalibration exists; an automatic safeguard does not.
  • Confidence is evidence strength, not probability. A score of 90 does not mean ninety percent likely. It means strong, sustained and clean relative to how this environment normally behaves.
What comes next

MAG‑NET — an anonymous observation network.

A single magnetometer in a single pocket is a personal instrument. Many of them, reporting independently, become something else: a way to tell a local quirk from a real regional event.

The design constraint is stated up front. Contributions carry approximate location and a rotating anonymous identifier. Nothing published identifies a person or traces a movement. Individual spikes are not headlines — the network's value is in agreement between independent observers, not in amplifying every reading.

Now

Personal instrument

Detection, event recording with waveform signatures, mapping, confidence scoring, diagnostics, export.

Phase 2

Field characterisation

The tuning campaign the constants have been waiting for: measured noise floors, adaptive thresholds, motion discrimination, real detection ranges.

Phase 3

The network

Anonymous, approximate, aggregated contribution. Multi-device corroboration as the basis for elevating confidence.

Phase 4

Maps, replay, analysis

Regional activity over time, recurrence at known sites, and the tools to look back through it.