Evidence

What we built, and what was measured on it.

Every entry below is an internal reference build: something we built ourselves to prove a capability works, not a customer deployment. Where a figure exists it is published with what it was computed over, and with what it does not establish in the same card. Where no figure exists the card says so rather than reaching for one.

We do not name a counterparty on this page. Where work has been done with another organisation, it stays off this list until they have agreed in writing to be named, which is slower than the alternative and the only version of this page worth reading.

Dated records

Three builds, each with its date and its limits.

Brighton and Hove roadside pass

14 August 2026Internal reference build

What was built: One survey pass of roughly a kilometre of Brighton seafront and the streets around it, driven on a single day. 2,008 frames, 4,146 positioned features and 3,411 carriageway defect reads, delivered as one layer per asset class in British National Grid with a data dictionary and stable feature identifiers, and browsable in the capture viewer with the photograph behind every value.
What was measured: Legible-value precision of 98.0%, 245 of 250, on a seeded reproducible stratified sample, with a 95% confidence interval of 95.4% to 99.1%.
What it does not establish: That figure measures whether values we claimed to read were read correctly, and nothing else. It is not a recall figure, so it says nothing about what the pass missed, and positional accuracy has not yet been measured against survey-grade ground truth. Coverage is what was driven: there is no claim to national coverage here.

How the pass and the sample work

EV charging app

20 July 2026Internal reference build

What was built: A nationwide EV charging companion built by our agents against the MapMap platform: a branded vector basemap, clustered site pins, turn-by-turn routing to a chosen charger and a reachable-range isochrone ring. 909 UK charging sites. Built in under a day.
What was measured: What was measured is the build, not the driving: the map, the routing and the range ring are live calls to the hosted gateway, and the app was standing up and working inside one working day.
What it does not establish: It is not a customer deployment and it is not affiliated with any charging network. Charger availability and pricing come from a dated snapshot rather than an operator feed, and the charge session is simulated, so nothing here evidences live third-party data or a commercial integration.

Open the running app

Low-bridge capture app

21 July 2026Internal reference build

What was built: A crowdsourcing app for the UK bridge-strike problem, built by our agents in under a day: four thousand official low-bridge records seeded as quest pins, in-app camera capture with evidence hashing, a moderation console with an asymmetric evidence bar, and a route-impact view putting a height-blind car route beside a dimensioned truck profile on the same bridge.
What was measured: What was measured is the build and the routing behaviour it demonstrates: the truck profile refuses a bridge the car profile passes under, computed by the same gateway that serves the platform.
What it does not establish: The seeded bridge records are official data loaded as a starting point, not our own survey, and none of the seeded heights has been verified by capture. No contributor-submitted value is evidenced here, and no accuracy figure is claimed for the seeded set.

Open the running app

How to read this page

A build is not a deployment, and a measurement is not a guarantee.

Labelled

Internal means internal

We built it, we paid for it, and nobody bought it. That is worth something as proof that a capability exists and worth nothing as proof that someone wanted it. The distinction is the reason each card carries the label rather than leaving it to the reader.

Dated

The date is the work

Each date is when the work was done, not when this page was written. A survey pass describes the street on the day it was driven, and a build describes the platform as it was on the day it was built. Both move, and an undated claim hides that.

Bounded

Every figure has a limit beside it

Precision is not recall and neither is positional accuracy. Where we have measured one and not the others, the card says which, because a single percentage with no denominator beside it is the easiest number on a website to misread.

Want the same thing measured on your network?

Tell us what you hold and what you need evidenced, and we will scope it against a pass rather than a list.

Start the conversation How a pass works