Legible, Navigable, Writable: What Measuring a City Actually Costs

A planner at City Hall wants to know how many buildings in Kibera have more than one storey. It is not an unreasonable question. It bears on drainage, on fire access, on how many people are actually there. She cannot answer it, and neither can anyone else in the building. The question is not hard because the answer is secret. It is hard because nobody has ever written the answer down.

For most of the last two years this body of work has argued that African institutions are failing for want of operational data infrastructure rather than for want of integrity. This paper is the first one where I can put a number on what closing part of that gap actually costs — because we closed it, for one dimension, in one city, and the closing produced a result I did not expect, and a second result, about cost itself, that I only found by going looking for it after the fact.

0.3% measured to 73.6% measured — and one of the sources that said 'measured' was probably wrong.

1. Two claims under one title, kept separate

This paper mixes two different kinds of claim, kept separate below rather than left to blur together under one title.

What it argues from, for the measurement result (§2–§5). Direct, reproducible engineering: a described pipeline, a validation run against the only ground truth available in the dataset, and a second published study used to check the first result rather than take it on faith. The numbers in these sections are ours — produced by code that ran, against data whose provenance is stated — and I consider them the strongest claims in the paper.

What it argues from, for the cost result (§6). Published institutional history and a 2023 peer-reviewed study of global OpenStreetMap building-data completeness, read directly rather than summarised secondhand. This is new territory for this series — no source here was pre-verified by an earlier paper — and I treat it with the caution that implies: where a number is Kenya-specific and I could not verify it against a primary document, it is flagged CITATION-NEEDED in Appendix B rather than estimated.

What it does not argue from. No claim that the free pipeline described below is representative of what every data gap in African institutions would cost to close — building height for one city, riding on data someone else had already published, is a favourable case, and §9 says so plainly. No labour cost for building the pipeline itself; the paper reports what the data cost, not what the engineering time cost, and that gap is registered in Appendix B rather than guessed at.

What would refute it. Given in full in §12. The sharpest version: if a replication finds the 6.84-metre disagreement in §5 traces to a resampling or coordinate-system error in this paper's own pipeline rather than to the surveyed attribute, the paper's most interesting claim — that the source everyone treats as authoritative is probably the weaker one — is simply wrong, and I would withdraw it rather than retreat to a softer version.

2. Three states an institution's data can be in

It is useful to separate three things that get collapsed into the word "digital".

Legible means the thing can be seen. A building exists in a dataset with a location and a shape. Someone asking "what is there?" can get an answer.

Navigable means the thing can be traversed and interrogated in context. Not a row in a table but a place you can move through, adjacent to other places, at a scale a human can reason about.

Writable means the thing can be changed by the people it describes. A resident who can see that the model has their building at one storey when it has three can say so, and that correction lands somewhere that matters.

Almost all civic technology in this region stops at legible, calls itself a platform, and waits. The gap between legible and navigable is mostly engineering. The gap between navigable and writable is mostly institutional, which is why so little of it exists.

3. What the twin is now

The Nairobi twin now holds 1,384,206 real building footprints across the metro, streamed tile by tile so that the whole city is walkable without the whole city ever being in memory. Footprints come from Overture Maps, which merges Google's Open Buildings, Microsoft's ML-derived footprints, and OpenStreetMap.

Until recently every one of those buildings was rendered as a box, because the pipeline reduced each real footprint polygon to its minimum rotated rectangle before storing it. That was a defensible shortcut and a real loss: an L-shaped block, a courtyard building, or an angled plot boundary all became rectangles. The footprints now survive to the renderer, simplified only as far as needed to keep tiles small.

More importantly, the heights changed.

4. 0.3 percent to 73.6 percent

Overture carries a surveyed height for 4,231 of the 1.38 million buildings in the metro. That is 0.3 percent. For the remaining 99.7 percent, earlier versions of this twin estimated height from a heuristic: distance from the central business district, multiplied by a land-use factor.¹ Buildings near the core were assumed taller. Nothing about any individual building was observed. The estimate was labelled as an estimate and colour-coded by confidence, which is the honest minimum, but it was still fundamentally a guess dressed in a gradient.

Google's Open Buildings 2.5D Temporal dataset changed what was possible. It provides building heights derived from satellite imagery across Africa, South Asia, South-East Asia and Latin America, annually from 2016 to 2023, at roughly 4-metre effective resolution, under CC BY 4.0 and ODbL.

The twenty-five source rasters covering greater Nairobi total 9.65 GB at their native half-metre tiling. We did not download them. They are Cloud-Optimized GeoTIFFs, so the pipeline reads only decimated windows over HTTP range requests and mosaics them into a single 16 MB raster at the dataset's true effective resolution. Then it samples that raster inside each building's real footprint polygon and takes the median.

The estimator was chosen by measurement rather than by taste. Validated against the buildings where Overture carries a surveyed height, taking the median of the pixels inside the footprint beat both the 90th percentile and the maximum. Using the maximum sounds appealing — you want the roof, not the ground — but it is noise-sensitive and it was the worst of the three.

The result:

Nearly three quarters of the city now stands at a height that something actually looked at. The remaining quarter splits into two honest failures: 298,557 buildings read below two metres in the raster, which at 4-metre resolution mostly means structures too small for the grid to resolve, and 66,553 fall outside raster coverage entirely. Those are still guesses, still labelled as guesses.

Every dataset named in this section was free and openly licensed, and the pipeline that merges them fits in 16 megabytes of working memory. That fact becomes the subject of §6.

5. The result I did not expect

Having built the thing, I validated it against the only ground truth in the dataset — Overture's own surveyed heights — and got a mean absolute error of 6.84 metres.

My first instinct was to write that up as a finding about the raster. Google publishes a 1.5-metre mean absolute error for this dataset, with the caveat, in their own documentation, that it was evaluated in the Global North only. Here was a Nairobi number four and a half times worse. It is a tidy narrative: the Global South gets the worse model again.

It is also wrong, and it took a citation to catch it.

Watson and Elliott's 2025 paper in Scientific Reports evaluated building-height datasets against ICESat-2 satellite laser altimetry — a reference with roughly half a metre of uncertainty — for Nairobi, Kathmandu and Quito.² For Nairobi it puts Open Buildings Temporal at 2.49 metres mean absolute error, better than the authors' own deep-learning model at 7.02 metres, with a tendency to overpredict by around 2.2 metres.

So the raster is roughly two and a half metres off against laser altimetry, and roughly seven metres off against Overture's surveyed heights. Both cannot be true of the raster. The difference is the ground truth.

I checked the obvious mechanical explanation first. My mosaic resampled half-metre pixels onto a 4-metre grid by averaging, which would dilute roof heights with surrounding ground and could plausibly account for the gap. So I rebuilt the entire metro mosaic using nearest-neighbour sampling instead and re-ran the validation. The results were identical to two decimal places. That rules out resampling, and incidentally confirms what the dataset documentation implies: the half-metre rasters are upsampled from a 4-metre native signal, so the two resampling strategies are reading the same underlying numbers.

Which leaves the conclusion I did not want. Two sources that both describe themselves as measured disagree by 6.84 metres, and the one with independent laser-altimetry validation is the satellite raster, not the surveyed attribute. Much of that gap is more likely error in Overture's height field — which inherits from OpenStreetMap contributors, from modelled values, and from sources of varying provenance — than in the imagery.

The limit of this finding is worth flagging honestly rather than letting the sentence above stand unqualified: Watson and Elliott validated three cities, and Nairobi's result is one of them. It is independent, peer-reviewed, and the best ground truth available to me, but it is not a general audit of Overture's surveyed-height field across the world, and §9 takes this up directly as an objection rather than leaving it as a footnote.

6. What the alternative would have cost

Here is the question the title of this paper promises and the sections above have not yet answered: what would it have cost, institutionally, to get Nairobi's building heights the conventional way — a government survey, properly budgeted, rather than a free pipeline stitched together from other people's published data?

Kenya has already tried to build the institution that would do this, and the attempt has a documented, verifiable history. A 2017 peer-reviewed study by Mwange, Mulaku and Siriba traces the Kenya National Spatial Data Infrastructure (KNSDI) from its launch in the early 2000s, through a period of real momentum funded by the Government of Japan via JICA, to the end of that funding in 2009, after which activity dropped sharply.³ By the authors' own 2014 measure, Kenya's SDI-readiness index stood at 0.56 — low enough, in their words, that "significant effort is still required."³ By 2016 the national geoportal, meant to be the public face of the infrastructure, was not even accessible.³

The funding picture is the part that matters most for this paper. A review of Kenya's 2016/2017 programme-based national budget found no direct line item for the KNSDI at all — the authors report it had apparently been absorbed into unrelated land-reform and survey programmes, with no separate accounting.³ Informal sources at the Survey of Kenya put the real figure at around KES 300 million (roughly US$3 million) for that year — enough, the authors write, "for the secretariat" but "inadequate for the overall SDI development."³ Their own costed roadmap for actually relaunching the KNSDI put the price at approximately KES 7 billion (about US$70 million, or 0.1% of GDP in 2016) over a five-year period — more than twenty times the money that year's budget actually provided.³ As far as I have been able to establish, no public record exists of that relaunch budget being funded; §11 registers this as an open question rather than an assumed no.

That is a budget shortfall in exactly the shape this series has already documented for a different institution: in the compliance-layer paper, the state was funded to process applications, not to reduce the cost of meeting them, and the gap between the two was structural rather than careless. The same shape appears here. A national mapping institution is a real line item that briefly existed, that a donor funded for six years, and that nobody's budget has reached since — not because measuring a country is unusually hard, but because, as that earlier paper argued for a different domain, no single party's budget is sized to carry a cost whose benefit is everyone's and whose bill is nobody's in particular.

The 2023 Nature Communications study on global OpenStreetMap building completeness, already used earlier in this series for a different argument,⁴ makes the same point at a different scale. Herfort, Lautenbach, Porto de Albuquerque, Anderson and Zipf report that building data "are usually maintained by national statistics offices," which are "underfunded by an estimated gap of $1 billion USD globally," and that this is why baseline geospatial data "are often not accessible, not up-to-date or not available in standard formats."⁴ Sub-Saharan Africa's building-completeness figure in that same study — 30%, "slightly higher than the global mean" of 24% — looks, at first glance, like a counterexample to the funding story.⁴ It is not. The same paper reports that more than half of all building edits in Sub-Saharan Africa — 51.1% — came from organised humanitarian mapping campaigns, not from ongoing institutional budgets.⁴ The region's relatively good number is a product of concentrated, externally funded, event-driven attention — the response to a flood, a disease outbreak, a specific crisis — not evidence that the underlying institutional funding gap closed. It is the data equivalent of the humanitarian mapper standing in for the national statistics office the KNSDI paper shows Kenya could not keep funded.

Put the two findings together and the twin's own result in §4 looks different than it did before this section. 73.6% of Nairobi's buildings now carry a height someone actually measured, achieved with a 16-megabyte pipeline built on entirely free, donated data. The KNSDI paper prices the institutional alternative — Kenya building and sustaining the capacity to do this kind of measurement itself, as a matter of ongoing government infrastructure — at roughly $70 million over five years, a sum the historical record suggests the country has not found twice in two decades of trying. The free pipeline did not succeed because measuring a city is cheap. It succeeded because it did not ask Kenya's budget to pay for it, and rode instead on data that Google and thousands of unpaid and crisis-funded OpenStreetMap contributors had already produced for other reasons. That is a real achievement and also, on reflection, a fragile one: §9 returns to exactly how fragile.

7. What this does to the design

The twin ranks height sources by confidence: surveyed beats imagery beats heuristic. That ordering was written before the result in §5 and it is now only partly defensible.

It remains right in one specific place. The raster caps at 100 metres. Nairobi has towers well past that — Overture has Times Tower at 150 metres — so for the tall core, the surveyed value is the only one that can be correct. Everywhere else, the ordering rests on an assumption that "surveyed" means "better", which this measurement does not support.

I have not changed the precedence yet, because changing it on one comparison against a ground truth of unknown provenance would be trading one unexamined assumption for another. It is recorded in the pipeline as a known open question, and both figures — the 6.84-metre disagreement and the 2.49-metre validated error — ship inside the tile store's metadata, so the distinction travels with the data instead of living in someone's memory.

That is the actual lesson, and it generalises well beyond building heights.

8. Provenance is the product

An operating layer that stores values is not enough. An operating layer has to store where each value came from and how much to trust it, because the moment you have two sources you have a conflict, and the moment you have a conflict you need a rule, and the rule is only as good as the provenance you kept. The W3C's PROV data model formalises exactly this — entities, the activities that generated them, and the agents responsible — as a general-purpose standard for recording that kind of lineage, precisely because the problem recurs across far more domains than building heights.⁵

This is not an abstract concern for institutions in this region. A water utility reconciling a meter reading against a billing record, a health system reconciling a paper register against a digital entry, a county reconciling a land record against a survey — these are all the same problem. Two sources, both authoritative-sounding, disagreeing. Most institutional software resolves this by picking one silently. The twin now refuses to: every building in the model carries a visible marker of how its height was arrived at, and the disagreement between sources is recorded rather than smoothed away.

An institution that cannot say why it believes a number cannot defend that number, cannot improve it, and cannot delegate it. Provenance is not metadata hygiene. It is the precondition for an institution being able to reason about its own data at all.

9. Objections I take seriously

"Three cities is not enough to indict a global dataset's surveyed-height field." This is the most serious objection to §5, and I raised it myself there rather than wait for a reader to find it. Watson and Elliott validated Nairobi, Kathmandu and Quito — a real, independent, peer-reviewed check, but not a global audit of Overture's surveyed attribute. The honest position is that this paper's claim is well-supported for Nairobi and suggestive, not proven, as a general statement about the dataset. §12 states the direct test that would settle the general version.

"The free-data success story is not generalisable, and showcasing it risks making the funding gap in §6 look smaller than it is." I think this objection is largely correct and worth stating plainly rather than letting the paper's own enthusiasm obscure it. Building height, for one city, riding on two datasets a large company and a global volunteer community had already built for reasons unrelated to Nairobi's institutions, is a favourable case. It generalises to the extent that similar donated data exists for whatever the next dimension is — building footprints, road networks, land use — and does not generalise at all to the dimensions where no comparable donated dataset exists. §11 states this limitation directly.

"Humanitarian mapping is still real capacity, even if it is event-driven — why does the funding source matter if the data gets collected either way?" It matters because event-driven attention is not schedulable, not owned by the institution that needs the data, and not guaranteed to recur for the next crisis or the next city. §6's point is not that humanitarian-mapped data is low-quality; it is that a government cannot budget around a donor's or a volunteer community's attention the way it could budget around its own line item, which is precisely the KNSDI's documented failure mode.

"Flux has a commercial interest in this conclusion." Flux builds the Nairobi digital twin this paper reports on. That is a direct conflict and worth stating without softening. The mitigation available is in §6 itself: the paper's own argument is that the twin's headline result was achievable because it avoided needing Kenyan institutional budget, not because such budget exists or is forthcoming — which is a poor foundation for pitching Flux's work as a funded, scalable government service, and a more honest position than a sales narrative would adopt. §11 is explicit that this paper does not claim the approach generalises to data with no free equivalent.

10. What would falsify this

Stated as commitments, in decreasing order of how badly each would damage the paper's claims.

A replication finds the 6.84-metre disagreement in §5 traces to an error in this paper's own pipeline — a coordinate-system mismatch, a footprint misalignment, a unit error — rather than to Overture's surveyed attribute. That directly contradicts this paper's most interesting empirical claim, and I would withdraw it rather than retreat to "the sources merely disagree."

A broader validation, covering many more cities than Watson and Elliott's three, finds Overture's surveyed-height field performs as well as or better than the satellite raster on average. That would falsify the general version of §5's claim while leaving the Nairobi-specific result intact, since the two are separable claims.

Kenya's KNSDI, or an equivalent institution, is found to have been funded at or near the KES 7 billion relaunch figure since the 2017 study, with demonstrable building-height survey capacity resulting from it. That would falsify §6's claim that the institutional alternative remains unbuilt, and the "free pipeline succeeded by avoiding the budget problem" framing would need revision.

A comparable building-height result is achieved for a city with no equivalent free, donated dataset available, at a cost approaching what §6 prices the institutional alternative at. That would weaken the generalisability objection in §9 rather than this paper's core claims, but it is the test that would most directly answer the title's question for cases less favourable than Nairobi's.

11. Limitations, and what this paper does not claim

The free-pipeline result does not generalise to data dimensions with no comparable openly licensed dataset already published by someone else. Nairobi's building-height result rode on work Google and OpenStreetMap contributors had already done for reasons unrelated to this project.

It does not claim Overture's surveyed-height attribute is wrong in general, only that the specific, independently validated comparison available for Nairobi suggests the satellite raster is the more trustworthy of the two there. §9 states this boundary explicitly.

It does not claim a verified cost for the labour and engineering time this pipeline took to build, only for the data it used, which was free. Appendix B registers the engineering-cost gap rather than estimating it.

It does not claim the KNSDI relaunch budget in §6 was never funded — only that no public record of its funding could be found by the time of writing, which is a claim bounded by what I could establish, not an audited fact.

And it does not resolve the write-side problem named in §8's closing argument; that remains the next and harder direction.

12. What this leaves for the next direction

The twin is legible and navigable. It is not writable, and that is the whole remaining distance — a gap this series later checked against external digital-twin maturity frameworks rather than its own prior benchmarks, with the same conclusion reached independently: institutional data provenance, not rendering fidelity, is the thing actually missing.⁶

What writable would mean concretely: the resident who can see their building modelled at one storey when it has three can correct it, and that correction enters a queue that a real institution reads. Not a suggestion box. A record with provenance — who said so, when, on what basis — that sits alongside the satellite-derived value and the surveyed value as a third source with its own confidence tier, and that the merge rule has to reckon with.

This is where the argument in The Operating Workforce and The Personal Operating Layer has been heading. The reason it has not been built is not that the interface is hard. It is that a write layer implies an institution on the other end that has agreed to read it, and §6 of this paper is one account of why that institution's budget has historically not reached this far even for the read side.

I am not going to pretend that agreement exists. What this paper claims is narrower and, I think, more useful: we now know what the read side costs in data terms, and we have a documented account of why the institutional alternative has stalled twice in twenty years. Roughly three quarters of a city's building heights can be moved from guesswork to observation using free, openly licensed data and a pipeline that fits in 16 megabytes — a result that is real and worth having, and that exists specifically because it did not have to wait for the $70 million Kenya's own geospatial infrastructure has been short of since 2009.

The obstacle is on the write side, and it is institutional, in the same specific sense §6 makes precise for the read side: not that nobody wants it, but that nobody's budget is shaped to carry it. That is the next paper, and it will require an institution willing to be the first one to answer.

Appendix A — Evidence table

Sourced rows are cited in the footnotes. The unverified rows are measurements this argument needs and does not have; they are left empty rather than estimated.

Row 13 is the one most worth someone checking against Kenya's current national budget before this paper's institutional story is relied on as still current — the 2017 study is the most recent public accounting I could find, and I could not verify whether it has been updated.

Appendix B — Citation register

CITATION-NEEDED — current (2026) KNSDI funding status. Appendix A row 13. The Mwange, Mulaku and Siriba figures are from a 2017 paper reporting 2016/2017 budget data. I searched for a more recent public accounting of KNSDI funding and did not find one I could verify; the institutional story in §6 should be read as "as of the most recent public record I could locate," not as current-year fact.

CITATION-NEEDED — engineering and labour cost of this pipeline. Appendix A row 12. §6 prices the data this project used, which was free. It does not price the engineering time spent building the pipeline that merges and validates that data, which is a real cost this paper does not have a verified number for.

CITATION-NEEDED — Overture surveyed-height error rate at scale. Appendix A row 15. §5's claim rests on Watson and Elliott's three-city validation. A broader study, across more cities and more of the surveyed-height field's actual provenance mix, would either strengthen or overturn the general version of this paper's central empirical claim.

CITATION-NEEDED — cost of an equivalent result without a free donated dataset. Appendix A row 14. §9 concedes this paper's result may not generalise past cases where comparable open data already exists; no verified cost estimate exists for the unfavourable case.