Glossary
The recurring concepts behind the writing. These are the same definitions that expand inline when you hover them in any essay.
- AI employee
- Software that is role-scoped, agentic, and accountable — it owns a task end to end, takes real actions, and logs every one to an auditable record.
- Bounding box
-
A rectangle on the map defined by four coordinates (west, south, east, north), used to say "give me everything inside this area" when querying geographic data.
Approximate by nature — fine for ratios and composition, misleading for absolute counts near the edges.
- Building footprint
-
The outline of a building as seen from directly above — a polygon on a map, with no height information. The basic unit of every open buildings dataset.
Footprints are what satellites and volunteers can actually produce at scale; everything else (height, use, condition) has to be measured separately or estimated.
- CBD
-
Central Business District — Nairobi's dense commercial core around KICC, Times Tower and Kenyatta Avenue; the best-mapped square kilometers in the country.
It's where every dataset is richest — which makes it a misleading place to judge data coverage from.
- Digital twin
-
A virtual model kept in live sync with one real physical thing — a term from NASA engineering, not 3D graphics. A city twin is only a twin if the data underneath is real and current.
Most things called digital twins are renderings. The rendering is the easy half.
- GTFS
-
General Transit Feed Specification — the open data format transit apps and Google Maps use to describe routes, stops and schedules.
Publishing Nairobi's matatu routes as GTFS (Digital Matatus, 2014) is the canonical local proof that opening civic data pays off downstream.
- GeoParquet
-
A columnar file format for geographic data — Parquet with geometry columns — designed to be queried straight from cloud storage without downloading whole datasets.
It is why pulling 1.27M Nairobi buildings took minutes, not a 178GB download.
- Height confidence tier
-
Our label for how much to trust each building's height: Measured (real data), High (settlement pattern), Medium (distance + land-use estimate), or Low (bare fallback).
An estimate that carries its own uncertainty can be refined later, tier by tier; one that hides it just looks like data.
- Informal settlement
-
A residential area developed outside formal planning and land-tenure systems — in Nairobi: Kibera, Mathare, Mukuru, Korogocho and others, housing a large share of the city.
They're systematically undercounted in volunteer map data, which means tools built on that data quietly leave them out.
- M-Pesa
- Kenya's mobile-money network — the canonical example of leapfrogging a generation of infrastructure rather than building a faster version of it.
- Matatu
-
Nairobi's privately-operated minibus — the transit system roughly 3.5 million people ride daily, run as thousands of small businesses rather than one public agency.
Any Nairobi civic model that ignores matatus ignores how the city actually moves.
- Once-only principle
-
The e-government rule that a citizen should never have to give the state the same fact twice — recorded once, a fact flows to every system that legitimately needs it.
It converts paperwork from five parallel processes into one event with consequences — the felt difference between digitized forms and digital government.
- Overpass API
-
OpenStreetMap's public query interface — ask it for every tagged building or road inside a bounding box and it returns the live map data.
It's how the v1 CBD scene was built, and it's free — but it only returns what volunteers have traced.
- Overture Maps
-
An open map-data project (Linux Foundation, backed by Amazon, Meta, Microsoft and TomTom) that merges OpenStreetMap, Google Open Buildings and Microsoft ML footprints into one deduplicated, freely usable dataset.
Using the merged layer instead of any single source is why Mathare's buildings appear on our map at all.
- ProcureBee
- Flux's first AI employee — an autonomous procurement agent, scoped to the role where institutions lose the most and can verify the most.
- Riparian land
-
The legally-protected strip of land along a river or stream where construction is restricted because the river needs room to flood.
Construction on riparian land is a named cause of Nairobi's 2024 flood deaths — and exactly the kind of violation a live city model makes visible before the rains, not after.
- Tile streaming
-
Serving a large map in small grid squares ("tiles") fetched on demand as the viewer pans, instead of loading everything at once. How every web map since Google Maps has worked.
It is the difference between a 68MB dataset being instant and being unusable in a browser.
- X-Road
-
Estonia's national data-exchange layer (launched 2001): each institution keeps its own records and answers queries from others peer-to-peer, with every query logged and visible to the citizen concerned.
It is the reference architecture for the invisible half of a real digital twin — institutions made mutually legible without a central superdatabase.
- first-fill
- Filling a role an institution never managed to staff at all — as opposed to replacing an existing worker. The African case, not the Western one.
- non-revenue water
- Water a utility produces but never bills or collects on — the clearest measurable symptom of an institution operating blind.
- operating blind
- An institution that lacks the feedback loops to observe what it did yesterday or is doing today — it cannot correct its own drift.
- operating workforce
- The next layer after visibility: AI employees that do an institution's routine, rules-bound work — not just show it.
- personal operating layer
- The missing-layer thesis pointed at an individual life: an intelligent, continuous record of what you're trying to do and how it's going — living in your messages, not an app to tend.
- the missing layer
- Flux's founding thesis: African institutions lack the operational data layer that would let them see their own functioning in real time.
- vendor capture
- When a single supplier ends up holding operational control of public infrastructure the institution cannot do without — monopoly produced through procurement.