The 8 fields in a position report, and the 2 that make it trustworthy

A position report — the ping — is one message from a vehicle saying where it was, when, and in what state. The coordinate pair is the payload. Everything useful sits in the 8 fields around it: device time, server receipt time, fix accuracy, satellite count or fix type, heading, speed, odometer and ignition state. Two of those 8, device time and fix accuracy, decide whether the rest is evidence.

A latitude and longitude alone cannot be filtered, ordered reliably or defended. You cannot tell whether the fix is 20 seconds old or 4 hours old, accurate to 5 metres or 2 kilometres, taken moving or parked. Live maps, arrival alerts and mileage claims rest on the qualifying fields, not the coordinates.

The 8 fields, and where each one comes from

Provenance is the thing to hold on to: some fields are measured by the device, some added by a server that never saw the vehicle, some sent only by certain vendors.

Ignition varies most between installs: 4 mechanisms produce it, and they disagree in ways only visible in the data — see what an ignition event actually proves.

Two clocks, two different questions

Device time is when the vehicle was there. Receipt time is when you found out. Through tunnels, underground bays and rural notspots those diverge constantly: a unit buffers fixes while offline, then dumps 40 of them within 1 second of reconnecting.

• Order the timeline by device time. Ordering by arrival produces a trail that doubles back whenever a buffer flushes.
• Measure staleness by receipt time. Answering 'last heard from' with the device clock hides a broken ingest path, which is the ambiguity behind vehicles frozen on the live fleet map.

Accuracy is what turns a position into evidence

Accuracy keeps 2 downstream systems honest. It sets the smallest zone whose crossings still mean something, which is why sizing follows fix quality in geofence triggers, entry, exit and dwell. It is also the first input to snapping fixes onto the road the vehicle was really on: a matcher weighting every fix equally anchors a route to a 400-metre outlier.

Speed and heading: reported, or reconstructed

A receiver can report speed directly from the Doppler shift on the signals it tracks: independent of position error, usable at walking pace. Speed reconstructed from the distance between 2 fixes over elapsed time inherits the error of both — with 30-metre scatter and a 10-second interval, a parked vehicle is assigned 10 km/h. Heading fails harder: many devices emit 0 when stationary, so 0 on a parked vehicle means 'unknown', not 'facing north'. Record which source each field came from before the series becomes the breadcrumb trail the vehicle actually drove.

The minimum contract to accept from a feed

1. Require both timestamps in UTC, device clock stored as sent. Correct at read time, never at write time.
2. Require an accuracy figure and record its units. Metres for 1 vendor and a unitless dilution figure for another is worse than none: both look plausible.
3. Store an explicit null, not a 0, for anything the device did not send, and keep the raw payload beside the parsed record.
4. Reject nothing at ingest — flag low-quality fixes and keep them. A silent drop rule cannot be audited when a mileage claim is tested against what the load tender committed you to.

A coordinate with no trusted time and no accuracy to filter on is a rumour about where a vehicle was, formatted to 6 decimal places.

How often a report is sent, and how trips are cut from the stream, are separate decisions sitting alongside this definition in telematics and the live vehicle data plane. Building a state model on such a feed is what we do under AI agents and workflow automation, with the operational realities in AI agents in production and the wider picture in our logistics and mobility work.

• telematics
• position report
• GPS
• fleet data