What is Geofence hysteresis?

A thermostat set to 20°C does not switch the heating on and off every few seconds as the temperature wobbles around that number. It heats until 20.5°C and does not restart until the room drops to 19.5°C. That gap between the switch-on and switch-off points is hysteresis, and it exists purely to stop rapid oscillation around a single threshold.

Geofences have exactly the same problem and the same fix.

The flapping problem

A single boundary line means a single threshold. A device sitting near it, with positions scattered by GPS drift, produces readings on both sides. The system sees inside, outside, inside, outside — a state change every few minutes from a device that has not moved.

This is called flapping, borrowed from network monitoring where a link that repeatedly goes up and down causes the same trouble.

The geometric fix: two boundaries

Hysteresis replaces the single line with two:

• An exit boundary, slightly outside the nominal zone
• An entry boundary, slightly inside it

A device counts as having left only once it crosses the outer line, and counts as having returned only once it crosses the inner one. The band between them is a dead zone where no state change happens at all.

For this to work, the band has to be wider than the position error. With a device accurate to 20 metres, a 10-metre band achieves nothing — drift crosses both lines. A 50-metre band absorbs it.

The cost is responsiveness. A device that genuinely leaves is not reported until it has travelled past the outer boundary, so a wide band means later alerts. That trade — false alerts against alert latency — is the real design decision, and it depends on the use case. A child leaving school warrants a tight, responsive zone with more noise tolerance elsewhere. A trailer in a yard can afford a wide, quiet band.

The state-machine fix

There is a second approach that solves the same problem without touching the geometry: make the notification stateful rather than the boundary.

Instead of two thresholds, keep one boundary and track whether you have already alerted. Suppress repeats while the alert is open, and only re-arm once the device has genuinely returned. Geometric hysteresis prevents the state from flipping; state-based suppression lets it flip but stops the flips from reaching the user.

What Geoblip does

Geoblip uses the state-machine approach as its primary defence.

Each device-zone pairing carries an alertStatus of ok or alerting, separate from the physical inside/outside flag. Exit notifications fire only on the ok → alerting transition. While the status is alerting, further apparent exits update the physical position but send nothing.

Re-arming requires a full alerting → ok → alerting cycle: the device has to properly re-enter the zone — which sends a recovery alert — before another exit can notify. Drift oscillating at the boundary cannot complete that cycle. Genuine movement can. The mechanism is described under alert debouncing.

Why this rather than dual boundaries? Two reasons. It keeps the zone the user drew as the zone the system uses, so the map shows the truth rather than a nominal line with invisible buffers either side. And it does not delay genuine exits — the first real crossing alerts immediately, with no band to traverse first.

Choosing your own buffer

Whichever approach a system uses, you still control the zone size, and that is the most effective lever available to you. Draw boundaries with slack:

• At least 50 metres of margin beyond the actual property line
• Include the car park, driveway, and anywhere a device might sit while “arrived”
• Never draw a zone smaller than your device’s accuracy — a 50-metre zone with 50-metre accuracy is noise generation, not monitoring

The geofence builder shows the real dimensions of what you draw, which makes it easy to check whether your margin is genuinely there.

Last reviewed 7 August 2026.