Questions about the measured record.
What we measure, how it is recorded, and what each product does with it. Δ10B+ labelled miles, Δ60M+ edge cases, Δ100M fresh miles every month.
What Empiric Earth is.
The entity, the promise behind it, and where the record comes from.
01What is Empiric Earth?
Empiric Earth is a physical-world measurement company. For ten years we have recorded what actually happens on the road — vehicles, assets, conditions, outcomes — and turned it into a queryable record of measured fact. The record holds Δ10B+ labelled miles and Δ60M+ edge cases, and grows by Δ100M fresh miles every month.
We are the company that operated as Nauto and Nexar. The measurement is the product: the intelligence is recorded from physical reality, not generated from a model of it.
02What does “measured, not imagined” mean?
It is the line that separates a recorded record from a generated one. Every mile in our record happened to a real person on a real road, and is labelled with what happened next. None of it is simulated, inferred from a proxy, or synthesised to fill a gap.
The distinction matters wherever the decision is expensive. A model trained on imagined driving learns a world that was never measured. An underwriter pricing imagined exposure prices a market that does not exist.
03Where does Empiric Earth’s data come from?
From connected cameras in vehicles driven by real people — fleet, consumer and OEM programmes — recording continuously on public roads. Ten years of it: Δ45PB of archive, Δ60M+ videos, and Δ100M fresh miles added every month.
Every event is labelled with the outcome that followed, which is the part a model or an underwriter can actually use. We compete with none of our buyers, so the record stays neutral.
Recorded, not generated.
How the measurement works, and why generated driving cannot stand in for it.
04What is a driving world model?
A driving world model is a learned model of how the road behaves over time — how vehicles, people and conditions actually evolve — rather than a classification of what a sensor saw in a single frame. It can be asked what happens next, and it can be tested against what did.
Its quality is set by what it was trained on. Trained on recorded driving with labelled outcomes, it learns the world. Trained on generated driving, it learns another model’s assumptions.
05Why can’t synthetic data replace recorded driving data?
Synthetic data can multiply a scenario you have already measured. It cannot produce one you have never seen, because a generator only recombines what was in its training set. The long tail — the Δ60M+ edge cases that decide outcomes — is exactly the part with no ground truth underneath it once it is generated.
It is also unfalsifiable. A recorded edge case comes with what happened next. A synthetic one comes with what a model assumed would happen next. Only one of those can validate a system.
06How is regional driving risk measured?
As events over exposure. Count the measured events of a given type in a region, divide by the measured driving exposure in that same region, and you get a rate that compares across places of different size and traffic volume: rate = events ÷ exposure.
Both terms have to be measured for the ratio to mean anything. Crash reports give you a partial numerator and no denominator at all. Because we record the driving itself, we measure both from the same source.
What the record is used for.
Different buyers, different interfaces, one measured record underneath.
07What is Empiric Earth Atlas?
Atlas is the interface to the recorded edge cases — the scenarios that stall an AV or ADAS programme. Teams search the record for a specific situation, pull the real footage, and validate against driving that actually happened rather than driving they generated.
It is built for AV and ADAS validation teams, and it draws on the same Δ10B+ mile record as everything else we publish.
08What is BADAS?
BADAS is our driving model family for OEMs and tier-1 suppliers: perception and safety behaviour trained on the recorded record, so a manufacturer can ship mandated safety functions without building the measurement stack underneath them.
BADAS-Open is the published open version, already in use by the research community. [TO CONFIRM — model sizes and open licence terms.]
09What is the Empiric Earth Risk Index?
The Risk Index ranks driving risk by region using measured events over measured exposure, drawn from Δ100,000 cameras recording public roads. It is a relative ranking, not an absolute probability: it tells you that one region carries more measured risk than another, and by how much, on a like-for-like basis.
We publish it that way deliberately. An absolute claim would need assumptions we have not measured. A relative ranking is what the measurement actually supports.
10How do cities get up-to-date road data?
Most cities work from data collected on a survey cycle, so the record is out of date the day after it is captured. CityStream replaces the cycle with continuous measurement: signs, work zones, surface condition and road change, observed as vehicles pass and refreshed as they keep passing.
A city or DOT can search an area, keep a corridor or work zone under watch, and be told when it changes. The work-zone intelligence is delivered with IBM watsonx and is live with public-sector buyers today.
