Why your GPS trace lies, and what road-snapping actually fixes

A GNSS receiver in a vehicle does not know where it is. It knows how long signals took to arrive from four or more satellites, and it solves for a position that best fits those timings. In open country with a clear sky, that solution lands within a few metres. In a city street with tall buildings on both sides, signals arrive after bouncing off glass, and the receiver solves for a position that never happened, commonly ten to thirty metres from the truth, occasionally on the wrong road entirely.

This is not a hardware defect and it is not fixable by buying a better dongle. It is the physics of the measurement. Which means every telematics platform faces the same question: do you show the customer the measurement, or do you show them an interpretation of it?

Three places the error becomes your problem

Distance totals. A trace that wanders side to side is longer than the road it represents. Summed across a fleet over a quarter, raw-trace distance typically overstates real distance by two to five per cent, more in dense urban work. If you reimburse by the kilometre, or you are within sight of a lease excess-kilometre threshold, that is a real number with a dollar sign on it.

Speed evidence. Speed derived from position deltas inherits position error twice. Two consecutive noisy fixes a second apart can imply a speed the vehicle never reached. Any platform that raises a speeding alert from a single instantaneous sample will eventually accuse a driver of something they did not do, and that alert will be the last one they trust.

FBT and logbook records. A logbook entry is a legal record of a journey. If the start point is thirty metres away in the neighbouring street, the entry now says the driver began a business journey somewhere other than where they did. Nobody audits that in a good year. In a bad one, it is the detail that unravels a period.

What map-matching does

Map-matching, or road-snapping in everyday use, treats the trace as evidence rather than truth. Given a sequence of noisy positions and a road network, it finds the most probable path through that network: the route that explains the observations while obeying connectivity, one-way restrictions and plausible turn geometry. A position twenty metres into a paddock beside State Highway 1 is not evidence that the vehicle left the road; it is evidence that the vehicle was on the highway with a twenty-metre error.

Two things fall out of a good match. The first is distance measured along real road geometry instead of along a jitter path. The second, and the more valuable one, is that every part of the journey now has a road identity, and roads carry attributes. Classification. Direction. A posted speed limit.

Once a trip knows which road it was on, speed compliance stops being a guess about the driver and becomes a comparison between two known quantities.

Speed limits are the real prize

A speeding event needs three facts: how fast the vehicle was going, what the limit was, and for how long the gap persisted. Telematics gives you the first. Map-matching supplies the second, from the road segment the vehicle was actually on rather than from the limit of the road it was near. The third is a design decision, and it is where most platforms go wrong. A single sample over the limit is noise, while forty seconds at twelve kilometres per hour above it is a driving behaviour.

Vectis only raises an overspeed event when the gap is sustained, and grades its severity. It also records where each limit came from. When map data has no posted limit for a segment, the platform falls back to a default for the road's classification, and then to the organisation's own configured default, and keeps the source alongside the limit. An event that cannot name its own limit source should never be put in front of a driver as evidence.

What it does not fix

Map-matching is not magic, and vendors who imply otherwise are setting you up for an awkward meeting.

• The first minute or two. A cold-start receiver has the fewest satellites and the worst geometry exactly when the trip begins. Matching helps, but the start point can still land at the wrong end of a carpark. This is the one place where a premium refinement pass earns its cost, because it is where the errors change the meaning of the record rather than just the total.
• Parallel roads. A service lane beside a motorway, or a bus lane beside a general lane, can be within the error envelope of each other. A matcher will pick the more probable one, and it will sometimes be wrong.
• Missing map data. New subdivisions, private site roads and forestry tracks may not exist in the network at all. A trip through them will be matched to whatever is nearest, which is why we keep the raw trace and never overwrite it.
• Stationary noise. A parked vehicle with the ignition on still reports positions, and those positions still drift. Idle detection has to come from ignition and movement flags, not from position deltas.

How to interrogate your own platform

You do not need to take anyone's word for this. Four questions will tell you what you are looking at.

• Do you keep both traces? If the platform overwrites the raw trace with the matched one, the evidence chain is gone. You cannot audit a correction you cannot see.
• Which distance is on the report? Ask specifically whether the logbook and reimbursement figures use raw or matched distance. Then ask what the difference was, fleet-wide, last quarter.
• Where did the speed limit come from? Per event. If the answer is a single national default table, the platform is not doing segment-level compliance, whatever the dashboard says.
• Can it re-run? Map data improves. A platform that can re-enrich a historical trip after a map update is one that can correct itself; a platform that snapped once at ingest is frozen at the accuracy of the day it ran.

In Vectis, enrichment runs automatically on every completed trip, as a separate stage rather than inline with ingest. The raw trace is kept. Any trip, or any date range across a fleet, can be re-enriched without touching the underlying telematics records, so a map update that corrects an intersection can correct the trips that crossed it.

In the platform

Standard road intelligence, full-trip matching and per-segment speed limits, runs on every Vectis organisation at no per-trip cost. Premium refinement of the first and last two minutes is an enterprise add-on for fleets where cold-start accuracy carries legal weight. Read how the pipeline works.