[ RESOURCES · GLOSSARY ]

If a word carries a number,
it needs a definition.

This industry runs on terms that sound precise and are not. Every word below is defined the way we use it, so a claim built on it can be checked rather than taken on trust.

[ TERM / AS WE USE IT ]

Every word, as we use it

Real-world mileA mile of vehicle travel observed by a connected sensor and processed into the record. Not estimated, not simulated, not extrapolated from a sample.
Edge caseAn event that is rare in frequency, high in potential consequence, and poorly represented in average-distribution datasets. Because the boundary is a judgement, we never use an edge-case count as a headline figure.
Compound eventTwo or more ordinary risk factors occurring together. Rarer than any single factor and typically more severe than the sum of them.
Outcome-attachedAn event stored with what happened next — loss, near miss, intervention, cost. Footage without this is a picture; with it, it is experience.
At-fault collisionA collision attributed to the operator of the insured vehicle under the customer’s own classification, stated per deployment rather than imposed by us.
Most-serious collisionTow-away, hospitalisation or fatality, where a customer’s reporting regime defines it that way. Never used interchangeably with total collisions.
Independent testingTesting designed and executed by a party with no commercial interest in the result. Our own benchmarks, however openly published, are never described this way.
Published benchmarkAn internal benchmark released with its category definitions attached, so the boundary is inspectable. Openness is not the same as independence.
Synthetic dataData generated by a model or simulator rather than recorded from the world. We do not train on it, and “zero synthetic” means exactly that.
De-identifiedFaces and licence plates obscured on the device before upload, with no person-level index and no journey reconstruction across events. We do not use this term to promise that no individual can be identified. Full description →
AnonymisedWe do not use this word. It implies a guarantee we have not measured. We describe what the product does to imagery instead, and we say where the measurement is missing.
Predictive fusionRoad, vehicle and driver state evaluated as one judgement rather than three independent signals with separate thresholds.
Alert fatigueThe point at which a driver stops responding to a system because it has been wrong too often. A safety failure, not a usability complaint.
Targeted collectionDirecting the live network to observe a specific condition, geography or behaviour, rather than waiting for the record to happen to contain it.
The delta  Δx = x1 − x0The measured difference between two observed states. Placed before a figure only where that figure has approved scope and method — never before an aspiration or an illustrative value.

Where a definition here conflicts with how the term is used elsewhere in the industry, we use this one and say so. Methodology →