# Roadside

MapMap Roadside is a positioned register of what stands beside the road on
the segments a highways team commissions: signs, road markings, street
nameplates, lighting columns, street furniture, kerbs, crossings, utility
covers and shopfronts. Each feature carries its position, how that position
was derived, the value read off it where the text was legible, and a link to
the dated photograph it was read from. A second, road-facing camera on the
same drive reads the carriageway surface. This page is the technical
summary; the [Roadside product page](/products/roadside) is the overview.

Roadside is a survey you commission, not an API you call: there is no
public endpoint that returns Roadside features, and no self-serve signup.
What you can open today without talking to anyone is the live viewer and
the data dictionary.

## What a highways team gets

**The register.** One layer per asset class, in British National Grid, with
stable feature identifiers so a later pass lines up against this one. Every
feature has the same fields (below), and a field that could not be filled
says so rather than being dropped.

**The photograph behind each value.** Every record links to the frame it
was read from. The frames stay on our side and are reached through the
per-feature evidence link, so your system stores the record and the link,
not the imagery. Hosted access to those links is included for three years
from delivery; the layers themselves are yours to keep whether or not you
renew it ([pricing](/products/roadside/pricing)).

**The carriageway.** A road-facing camera captures a geotagged frame of the
carriageway roughly every 4 metres of the drive. Cracking, worn markings,
patches and reinstatements, ironwork, edge fretting and inspectors' painted
marks are read as records, each with a class, a position, a confidence, a
graded severity and a link to its frame. Severity is a visual grade for
prioritising an inspection, never a measured depth.

**A method and QA statement per delivery:** how it was captured, how
positions were derived, what the human review sample covered, and where the
unknowns are. Your term maintenance contractor and consultants can use the
data for your scheme by notification.

## Faces and number plates

Faces and number plates are detected so that they can be obscured, and they
are obscured before a frame is shared, delivered or published: before
anything leaves us. Detection is not identification; nothing is matched to
a person, a record or another pass. The obscuring on a delivered set is
audited, and the audit date is recorded beside the capture date, so a
published sample carries both (the public demo: captured 14 August 2026,
blur audit cleared 25 August 2026).

Blurring does not make a frame anonymous, and we do not say it does:
blurred frames are still personal data, and raw and blurred frames are held
separately. What is captured, the lawful basis, and how anyone in a frame
can ask for it to be removed are in the
[Roadside imagery notice](/legal/roadside-imagery).

## The live viewer and the sample pack

- **The viewer:** [roadside.mapmap.ai/demo](https://roadside.mapmap.ai/demo/)
  serves the seafront kilometre of the Brighton and Hove pass of 14 August
  2026, openly and with no login: 4,146 positioned features, 3,411
  carriageway defect reads and 2,008 blurred photographs, each with its
  evidence link working. Those counts describe the extract, not the full
  54.9 km pass. Click a feature and it opens the frame it was read from.
- **The sample pack:** [/roadside/sample-pack](/roadside/sample-pack)
  publishes the data dictionary in full. The sample layer itself, one asset
  class from the same pass as GeoJSON with the dictionary beside it, is
  sent on request from that page.
- **Routing on the surface reads:**
  [/demo/roadside-cycle](/demo/roadside-cycle) scores two bicycle routes
  between the same points against the carriageway reads from the Brighton
  pass ([demos](/docs/demos#roadside)).

## Formats and exports

| What | Form |
| --- | --- |
| Delivered layers | One layer per asset class, British National Grid, schema `roadside-products-v4`, stable `object_id` per feature |
| Sample layer | One asset class as GeoJSON, sent on request |
| Evidence | A per-feature link to the frame, held and served by us |
| Documentation | The data dictionary, and a method and QA statement per delivery |

Every feature carries the same fields, grouped as the dictionary groups
them:

- **Identity and provenance:** `object_id`, `category`, `category_label`,
  `subtype`, `street`, `drive`, `captured`.
- **What was read:** `value` (null where nothing legible was read),
  `legible`, `detail`, `conf_max_pct`, `conf_mean_pct`.
- **How it was placed:** `pos_method`, `bearing_deg_true`,
  `surface_feature`, `observations`, `obs_positioned`, `obs_bearing_only`,
  `frame_first`, `frame_last`, `provenance`.

`pos_method` is one of `triangulated`, `ground_cluster`, `linear_run`,
`merged`, `value_merged`, `single_finding` or `unlocated`, and it is on
every feature: a feature that could not be placed says `unlocated` rather
than sitting somewhere plausible. The
[sample pack](/roadside/sample-pack) defines every field.

The public viewer does not export: its GeoJSON and CSV export controls are
turned off on the public deployment, and the layers are not in the public
extract at all. Layers reach you as a delivery, or as the sample layer on
request.

## What the numbers mean

The published reading figure is legible-value precision of 98.0% (245 of
250) on a seeded, reproducible stratified sample from the first surveyed
town, with a 95% confidence interval of 95.4% to 99.1%. It measures whether
values we claimed to read were read correctly, and nothing else.
`conf_max_pct` and `conf_mean_pct` say how sure a read was, not how right
it was.

## What is not offered yet

- **Positional accuracy is not measured.** No figure has been measured
  against survey-grade ground truth and none is published. Treat positions
  as estimates labelled by `pos_method`.
- **No recall figure.** Precision says nothing about what a pass did not
  see.
- **Change reporting between passes is in build.** Until it ships, a
  re-survey is delivered as a fresh dated pass, not an automated
  difference.
- **No statutory condition survey.** Roadside is asset inventory and
  evidence; it is not a carriageway condition survey under PAS 2161, and we
  are not on the approved list. Open-pothole detection is undemonstrated.
- **No coverage claim.** We do not sell by area covered: what exists is
  what has been driven, and each pass says where it went.
- **No published imagery by default.** Publication of any imagery is
  decided in the data protection assessment and is not offered by default.
- **No public API or self-serve access.** Roadside is commissioned; start
  from the [product page](/products/roadside) or
  [talk to us](/contact?topic=sales&products=roadside).

## Related

- [Roadside product page](/products/roadside) and
  [Roadside pricing](/products/roadside/pricing)
- [Roadside sample pack and data dictionary](/roadside/sample-pack)
- [Roadside imagery notice](/legal/roadside-imagery)
- [Live demos](/docs/demos#roadside)