Stage Three of Eight

Seven papers in, we have never once asked whether KaNairo is a digital twin by anyone's definition but our own.

That sentence should sting a little, because we set out, in WP27, not to build a pretty rendering — we set out, explicitly, invoking Michael Grieves' 2002 concept and NASA's John Vickers' 2010 name for it, to build something that behaved like a spacecraft twin: live, queryable, wrong the moment it went stale.¹ Everything since has been in service of that. WP30 built a way to estimate building heights honestly instead of guessing silently. WP29 found and named the settlements OpenStreetMap simply wasn't seeing. WP28 rebuilt the whole pipeline so it could run at 1.27 million buildings instead of 1,000. WP32 got satellite-measured heights from 0.3% of the city to 73.6% and caught two "measured" sources disagreeing by close to seven metres in the process. WP37 designed, but has not shipped, a way for residents to write to the record without the write silently overwriting someone else's claim.

That is a real and unusually well-documented body of engineering. It is also, on its own, not evidence of anything relative to the word "twin" — because we invented the test we've been grading ourselves against. Every one of those papers measured KaNairo against KaNairo's last version. None of them measured it against a definition somebody else, with no stake in Flux's outcome, actually wrote down and published.

That test exists. Several tests exist. The UK ran a five-year, government-funded programme that produced a public, citable definition of what a national digital twin has to do to earn the name.² Researchers who study nothing but this question have built maturity ladders with named stages, and have gone out of their way to specify what a plain 3D city model is not allowed to claim.³ The manufacturing world has an ISO standard for what "digital twin" means when the thing being twinned is a machine, and it is worth checking whether — and how uncomfortably — that standard reaches into buildings.⁴ Two other cities, Singapore and Dubai, have put real money and real years into systems they call twins, and there is enough public record to ask what they had that we don't.⁵

This paper runs KaNairo against those tests. Not against a softened version of them, and not against a version where "well, but Nairobi is different" gets to excuse every gap. Some gaps genuinely are structural and Nairobi-specific — that turns out to be the more interesting finding, not an excuse. Others are just gaps.

A city-scale render sitting at stage three of eight is not a lesser twin. It is not a twin.

1. Scope, method, and standing

This paper argues one claim: measured against published, external digital-twin maturity and governance frameworks — not against KaNairo's own prior benchmarks — KaNairo currently sits at the "3D static/dynamic" tier of those frameworks, not at the tiers those same frameworks reserve for the word "twin," and the specific, nameable reason is a lack of institutional-grade, machine-interoperable data provenance, not a lack of rendering fidelity.

What this argues from: the numbers already published and defended in WP27 through WP32, and WP37 (I did not re-derive them; I am relying on the corpus's own prior verification work, so any error in those papers is inherited here); three external frameworks I fetched and read directly rather than took on secondhand description — the Gemini Principles paper itself,² the CITYSTEPS maturity model paper,³ and the Evans/Atkins five-stage spectrum;⁶ and direct, dated evidence about Kenya's and Nairobi's own institutional data infrastructure, gathered in September 2026.

What it does not argue: it does not claim KaNairo should stop shipping, should have been built differently from day one, or is a failure as a product. It does not claim any of the three frameworks used here is the correct definition of digital twin — the field doesn't have one settled definition, and a 2025 systematic review of over 15,000 publications found exactly that: no equilibrium, real disagreement even among specialists on which components are load-bearing.⁷ It does not claim Singapore's or Dubai's twins are complete, finished, or beyond criticism — the public record on both is thinner on governance than on capability, and I flag that thinness rather than paper over it.

What would refute this paper: if KaNairo, or a successor project, ships a live sensor feed with two-way write-back into the same building record a resident or institution edited — not a mockup, a shipped mechanism — that clears the stage-4/5 boundary on every framework used here and this paper's central claim is wrong going forward. If Nairobi City County's e-permitting system starts emitting structured, machine-readable building data (not PDFs) at any real coverage, the "structural, not engineering" framing of §6 is wrong and the gap becomes an ordinary backlog item. Either finding should retire this paper's thesis, not just soften it.

2. What KaNairo actually is, measured plainly

Strip the pipeline improvements the corpus already covered down to the current state, because the rest of this paper depends on getting that state right and not restating it in flattering terms.

KaNairo is a tiled, city-scale 3D building model of Nairobi: roughly 1.27 million building footprints, sourced by merging OpenStreetMap contributions with Overture Maps data, delivered as an 8,536-tile SQLite pipeline rather than a single monolithic file.⁸⁹ Of those buildings, 73.6% now carry a height derived from satellite raster imagery rather than a heuristic guess — up from 0.3% before that pipeline existed — and the remaining roughly quarter of the city relies on distance-based or small-structure assumptions, labeled as such.¹⁰ Coverage of footprints themselves is uneven in a way that tracks who has historically mapped what, not what's actually there: Kibera sits far higher than Mathare in human-contributed OSM density, a gap the corpus closed partly with satellite-derived footprint detection rather than waiting for volunteer attention to even out.¹¹ There is a designed-but-unshipped mechanism for letting a second party's claim about a building coexist with the first party's claim, rather than overwrite it — a provenance model, not a consensus or authority model, chosen because the other two required institutional or population preconditions Nairobi does not currently have.¹²

None of that is in dispute here, and none of it is small. Getting three-quarters of a city's buildings to a satellite-verified height, at this cost, is a real result, and this paper is not the place to relitigate it.

What KaNairo does not yet have, stated with the same plainness: no live sensor feed of any kind feeding into the model automatically. No mechanism by which an edit — resident, surveyor, or institution — writes back into a system anyone downstream queries; §6 of WP37 designed that mechanism, but "designed" and "shipped" are different words and the corpus has been careful, elsewhere, to insist on that distinction. No published schema mapping for a construction-industry data standard — nothing that would let a contractor's BIM software or a county planning system ingest a KaNairo building record without a bespoke, one-off translation. And no institutional co-signer: every byte in the pipeline traces back to a satellite imagery provider, a volunteer mapper, or Flux's own inference — not to Nairobi City County, the Kenya Space Agency, or the Survey of Kenya asserting "this record is ours and we stand behind it."

That is the model whose maturity this paper is placing.

3. Three definitions of "done," and what they actually require

Three frameworks, chosen because each is real, checkable, and independent of the others' authors, define "digital twin" — or its absence — as a specific technical threshold rather than a mood.

The Gemini Principles. In 2018, the UK's Centre for Digital Built Britain, backed by the Digital Framework Task Group and a government minister's foreword, published a definition intended to anchor a National Digital Twin programme: a digital twin is "a realistic digital representation of assets, processes or systems in the built or natural environment," and — critically — "what distinguishes a digital twin from any other digital model is its connection to the physical twin."² The paper is explicit that a static planning model with periodic, not live, data feedback still counts as a digital twin, so long as feedback into the physical asset exists at all; what it will not extend the word to is a model with no connection back to the thing it represents. It organizes nine "Gemini Principles" under three headings — Purpose (public good, value creation, insight), Trust (security, openness, quality), and Function (federation, curation, evolution) — and names, as a "key constraint," the fact that "few digital twins at present are connected or share data across organisations, sectors or geographies. Lack of interoperability is a key constraint."² That line is doing real work for this paper: the UK's own flagship digital-twin programme names interoperability, not rendering quality, as the thing actually blocking twins from being twins.

CITYSTEPS. Haraguchi, Funahashi, and Biljecki's 2024 maturity model, published in Technological Forecasting and Social Change, grades city-scale digital twins across eight stages: (1) preparation and planning, (2) 2D, (3) 3D static, (4) 3D dynamic, (5) dynamically integrated 3D, (6) real-time decision-making, (7) autonomous decision-making, and (8) real-time synchronization with autonomous implementation.³ The paper is explicit that stages 1 through 3 are, functionally, cartography and visualization — not yet a twin in any operative sense — and that the governance dimension (who curates, who can act on the model's outputs) is treated as a first-class axis alongside the technical one, not an afterthought bolted onto a finished system. A closely related 2023 review, from a different author group, makes the differentiation point even more bluntly as its title: city digital twins have to be told apart from 3D city models, because in practice the two get called by the same name constantly, and a systematic review of 41 papers on the subject found that 90% of studies reviewed sat at "initial to medium" maturity — 3D modelling, monitoring, and visualization — not the interactive or autonomous stages the word "twin" is usually used to imply.¹³

The Evans/Atkins spectrum. A 2019 IET whitepaper produced with the engineering firm Atkins, authored by Evans, Savian, Burns, and Cooper, proposed an earlier and more compact five-stage spectrum still widely cited in the field: reality capture, 2D/3D models, real-time data addition, two-way integration and interaction, and autonomous operations.⁶ Its stage 3 — the addition of live data — is the threshold this paper treats as the meaningful line between "model" and "twin," because it is the first stage at which the system can be wrong in a way that matters: a stale reading is now a failure mode, not just an incomplete dataset.

Three different research groups, three different years, three different institutional homes — CDBB/DFTG for Gemini, a Singapore-based urban analytics lab for CITYSTEPS, a UK engineering firm and professional body for Evans/Atkins — and all three independently draw the same line in roughly the same place: a static or slowly-updated 3D representation is a legitimate, useful thing, and none of them call it a twin on its own. All three also treat interoperability with an external data standard, and connection to an institutional source of truth, as part of what moves a project past that line — not a separate, optional concern.

There is no single canonical fourth framework this paper is obligated to check and didn't; the 2025 systematic-review paper cited above exists precisely because the field has not converged on one definition.⁷ What the three frameworks used here share, despite their independence, is the thing this paper is testing KaNairo against.

4. Running KaNairo against the scale, honestly

Placed on CITYSTEPS: KaNairo clears stage 2 (2D) without argument — it has been past flat maps since WP27. It clears stage 3 (3D static) on the strength of the tiled building-footprint-and-height pipeline described in §2. Stage 4 (3D dynamic) is the first genuinely contested placement, and the honest answer is partial: the height data has been updated in discrete, versioned passes as new satellite imagery and better inference methods became available — WP32's 0.3%-to-73.6% jump is itself an update event — but nothing in the published pipeline updates automatically or continuously the way "dynamic" implies in CITYSTEPS's own framing, where it describes a live-refreshing representation rather than a periodically re-run batch job. Stage 5 (dynamically integrated 3D) requires the write-back mechanism WP37 designed and did not ship; KaNairo does not clear it, and the corpus itself says so — WP37's own falsification conditions include "a city-scale twin ships multi-claim provenance and its consumers cannot use it," which presupposes the shipping hasn't happened yet. Stages 6 through 8 — real-time decision support, autonomous action, autonomous synchronization — are not remotely in scope for the current pipeline and I am not aware of any claim from Flux that they are.

Placed on the Evans/Atkins spectrum, the same shape appears with fewer rungs: KaNairo clears reality capture and 2D/3D modelling cleanly, and fails the "real-time data addition" stage, which is stage 3 of 5 — meaning on this framework KaNairo sits at roughly the 40% mark of the ladder, not the halfway-or-better position "digital twin" branding would suggest to an outside reader.

Placed against the Gemini Principles' Function trio — federation, curation, evolution — the picture sharpens rather than softens. Federation asks whether the twin connects, via secure shared data, to other twins or institutional systems; KaNairo does not, and §6 explains why that is not simply a to-do item. Curation asks whether there is clear ownership, governance, and regulation over who can change the record; WP37 built a provenance design for this and has not shipped it, so today the honest answer is that curation exists only for the read side (Flux curates what data enters the model) and not at all for the write side residents or institutions might use. Evolution — the capacity to adapt as technology and society change — is the one Gemini principle KaNairo has arguably satisfied better than most: the corpus's own six-paper arc of honest revision (0.3% to 73.6%, CBD to city-scale, single-source to merged) is closer to institutional learning than most twin projects can document about themselves.

That last point matters and should not be buried under the rest: on the axis of epistemic honesty about its own limits, which none of the three frameworks scores directly but all three implicitly reward, KaNairo is unusually strong. The corpus has never once claimed a heuristic was a measurement. That is not nothing. It is also not the same claim as "this is a digital twin," and this paper's job is to keep those two claims from blurring into each other.

5. The interoperability problem: the format nobody in this pipeline speaks

Every framework in §3 treats interoperability as part of the definition, not a bonus feature, and the built-environment world has a specific, real answer to what interoperability means in practice: Industry Foundation Classes (IFC), an open, vendor-neutral schema for building and infrastructure data, standardized internationally as ISO 16739-1:2024, maintained by buildingSMART International.¹⁴ IFC is how an architect's model reaches a contractor's bid software, and how a contractor's as-built model reaches a facility manager's maintenance system, across different vendors, without a bespoke one-off translator for every pair of tools.¹⁴ The forthcoming IFC5 revision is being designed explicitly to interoperate with GIS, IoT, and live sensor data — to be, in the standard body's own framing, twin-native rather than twin-adjacent.¹⁴

KaNairo does not speak IFC. Nothing in the published pipeline — OSM footprints, Overture Maps records, satellite-derived heights, SQLite tiles — is IFC data or maps cleanly onto IFC's schema, and I found no evidence in the corpus, nor anywhere public, that a KaNairo-to-IFC export has been attempted. This is not a criticism of an engineering choice; a tile-streamed SQLite pipeline optimized for web delivery was the right call for the problem WP28 solved, which was serving 1.27 million buildings to a browser, not exchanging design data with a construction firm. But it means that, right now, nothing built in construction software anywhere in Nairobi could read a KaNairo building record and do something useful with it, and nothing KaNairo ingests came from construction software either.

There is a second, related standard worth naming precisely because the field increasingly treats the two as needing to talk to each other: CityGML, an OGC standard for representing 3D city models, which sits closer to KaNairo's own domain than IFC does. Recent peer-reviewed work has specifically proposed semantic-graph methods for integrating IFC and CityGML data for "advanced 3D city model visualization" — which is a fair description of exactly the layer KaNairo occupies.¹⁵ That KaNairo predates and does not engage that integration work is not damning on its own; it is one more data point that the interoperability layer every framework in §3 treats as necessary has not yet been built into this pipeline, in either direction.

This is, to be fair to Flux, an extremely common gap — the same PMC-indexed 2024 paper on IFC/CityGML integration exists because almost nobody has solved this cleanly yet, and it is not solved in Singapore's or Dubai's public materials either, as far as I was able to verify.¹⁵ But "common" and "solved" are different words, and this paper's job is not to grade KaNairo on a curve against the rest of the field's failures.

6. Provenance and institutional sourcing: the gap that isn't really an engineering gap

This is the section that changes the shape of the argument, because the interoperability failure in §5 has a cause that is not primarily technical.

IFC and CityGML both assume there is an institutional actor — an architect's firm, a facility manager, a city planning department — producing structured, machine-readable data at the source, which downstream systems then exchange. Nairobi's own building-permit pipeline does not currently produce that. The county's e-permitting system, eDAMS, requires architects to submit building drawings as PDF attachments through an online portal; applicants track status by SMS and email, and the process is a genuine improvement in accessibility and speed over paper filing at a physical counter — but the artifact that emerges from an approved Nairobi building permit is a stamped PDF, not a structured IFC or CityGML record.¹⁶ There is, as far as I could establish, no requirement or mechanism anywhere in that pipeline for BIM-format submission.¹⁶ This is not a KaNairo problem. It is upstream of KaNairo entirely: even a hypothetical, perfectly-resourced version of this project could not source IFC-grade building data from Nairobi's own permitting institution today, because that institution does not produce it.

The geospatial-governance layer above individual buildings is moving, but only just, and the timing matters for this paper because it changes as this is being written. Kenya's National Spatial Data Infrastructure was first proposed years ago, coordinated by the Survey of Kenya with support from Japan's international cooperation agency, aimed at reducing land disputes and enabling spatial planning — but its own progress documentation describes an initiative still being relaunched and re-established, not a mature, operating national data backbone.¹⁷ More concretely and more recently: on 23 March 2026, following the adoption of the Malindi Declaration, the Kenya Space Agency stood up a National Earth Observation Data Council (NEODC) — with the British High Commission, the Research Institute for Innovation and Sustainability, and VIFFA Consult as partners, and the Kenya National Bureau of Statistics, the Survey of Kenya, the National Land Commission, and the Directorate of Resource Surveys and Remote Sensing as executive members — to coordinate a National Earth Observation Data Sharing Framework built around FAIR data principles (findable, accessible, interoperable, reusable), explicitly naming urban planning as one of its intended use cases.¹⁸ That is a real, dated, verifiable step toward exactly the kind of institutional data backbone the Gemini Principles' "federation" criterion requires. It is also six months old at time of writing, chaired by a coordinating council rather than operating as a live data service, and has no public building-level or Nairobi-specific output yet that I could find.

Put the two findings together and the shape of the gap becomes specific rather than vague, which is what this paper set out to establish: KaNairo cannot currently source federation-grade, institutionally-sourced, IFC-or-CityGML-interoperable building data from Nairobi, because (a) the county permitting process that would generate it at the source produces PDFs, not schema, and (b) the national framework that would coordinate and standardize geospatial data sharing across institutions is five months old and not yet operating at the scale or specificity a building-by-building city twin would need. Neither of those is something Flux's engineering team can fix by writing better code. Both are things that could plausibly change within the multi-year horizon this series has already committed to — the NEODC in particular is a live, moving target, and a future paper in this series should check back on it specifically rather than let this paper's snapshot calcify into a permanent verdict.

7. What Singapore and Dubai had before they used the word

Two comparisons, both handled carefully, because the public record on each is genuinely thinner than the marketing language surrounding them.

Virtual Singapore is co-led by the Singapore Land Authority, the National Research Foundation, and the Government Technology Agency, and reportedly cost in the region of $53 million (US) — a press-reported, order-of-magnitude figure, not a verified line-item budget.¹⁹ It was built on laser-scanning aircraft and ground vehicles capturing terrain and surface data directly, meaning its base layer was survey-grade capture commissioned by a national land authority with a statutory mapping mandate, not a merge of volunteer contributions and satellite inference.¹⁹ That is not a criticism of volunteer or satellite-derived data — WP30 and WP32 made a strong, correct case that labeled uncertainty beats waiting for a survey that will never come — but it is a real difference in what "institutional source" means: SLA is the institution, by law, that would have produced authoritative Singapore geospatial data with or without the digital-twin programme. Nairobi's building-permit authority produces PDFs.

Dubai's newer "Dubai Live" platform, launched at GITEX 2025 through a partnership between Dubai Municipality and the Dubai Land Department, is press-reported to span more than 195,000 buildings and 280,000 infrastructure assets — again, order-of-magnitude figures from launch coverage, not independently audited counts — and is explicitly framed as a real-time operations platform integrating AI and live monitoring across transport, land, and municipal systems, not a static rendering.²⁰ What Dubai had that this comparison can actually verify: a land department that already held authoritative parcel and building records as part of its normal legal function, and a municipality with the budget and mandate to wire sensors into that record directly. What this comparison cannot verify, and should not claim: how far Dubai Live actually is on any of the three §3 maturity scales, whether its governance model resembles federation/curation/evolution in the Gemini sense, or whether its "real-time" claims hold up to the same scrutiny this paper is applying to KaNairo. The press coverage found for this paper is capability-forward and governance-silent, which is itself worth naming rather than filling in with assumption.

The comparison this paper is entitled to make, carefully: both Singapore and Dubai built their twins on top of a pre-existing institutional actor whose day job was already producing authoritative, structured spatial or property records. Nairobi does not yet have that actor in the same form — its analogous institutions (Survey of Kenya, Nairobi City County, the nascent NEODC) are either under-resourced, mid-relaunch, or five months old. That is the actual asymmetry, and it is a resourcing-and-institutional-maturity gap between cities, not a competence gap between Flux and Dassault Systèmes or the Dubai Municipality's contractors.

8. Objections I take seriously

"You're grading against frameworks built for rich, small, data-saturated city-states — of course a Nairobi project fails." This is the strongest objection and I do not have a full answer to it. It is true that Singapore's land area and Dubai's budget make certain thresholds easier to clear regardless of institutional design. But the specific failure this paper identifies — no machine-readable output from the permitting process — is not a scale problem or a budget problem in the way "we can't afford enough satellites" would be; it is a format problem in a government IT system, and format problems are exactly the kind of gap that smaller, less-resourced digital-infrastructure efforts (Estonia's X-Road, cited in WP27, is the standard counter-example) have closed without Singapore-level budgets. So the objection correctly warns against treating the frameworks' thresholds as fair in an absolute sense, but does not, in my judgment, invalidate the specific §6 finding.

"Flux has a commercial interest in this conclusion looking either better or worse than it is." Flux builds and intends to sell KaNairo. That is a real conflict and it cuts in an uncomfortable direction for this paper specifically: an honest "not yet a twin, and here's why" is arguably worse for a sales conversation than the vaguer marketing claim the corpus has, to its credit, mostly avoided making. I do not think that protects this paper from the conflict — a sophisticated reader could argue that naming a structural, institutional cause (§6) rather than an engineering one is itself self-serving, because it shifts the "blame" for immaturity from Flux's roadmap onto Nairobi's government, which costs Flux nothing and may even read as a case for why Flux is needed. I take that version of the objection seriously enough to say plainly: nothing in §6 should be read as diminishing what Flux still has to build (§5's interoperability layer is squarely Flux's to build or not build) or as license to blame slow government IT for indefinitely deferring KaNairo's own maturity work.

"Maturity models like these are themselves marketing artifacts from consultancies (Atkins) or advocacy groups (CDBB), not neutral science." Fair, and worth stating outright: Evans/Atkins is an engineering-firm whitepaper, not a peer-reviewed study, and the Gemini Principles were produced by a government-convened task group with an explicit agenda to accelerate NDT adoption, not a disinterested academic exercise. That is why this paper leaned on CITYSTEPS and the 2025 Building and Environment definitional study as the more independent checks, and why §3 flagged, rather than hid, each framework's institutional origin. The three frameworks converging independently on the same rough line is the mitigating fact, not a claim that any one of them is neutral.

"This paper doesn't actually change what Flux should build next, so what is it for?" It changes what the corpus is allowed to claim in public. WP27 already committed to "not a screenshot" as the standard; this paper is the check on whether that commitment has been kept, stated in terms a reader outside Flux can verify against sources Flux didn't write.

9. Falsification conditions, in decreasing order of damage

1. KaNairo, or a documented successor, ships a live sensor-to-model-to-write-back loop — not a design, an operating mechanism, however small in scope (one utility feed, one neighborhood) — that clears CITYSTEPS stage 5 and the Evans/Atkins stage-3 threshold simultaneously. If this happens, this paper's central placement claim is wrong going forward, full stop, and should be retracted rather than merely updated.
2. Nairobi City County's eDAMS system, or a successor, begins emitting structured, IFC- or CityGML-compatible building data at any non-trivial coverage. If this happens, §6's "structural, not engineering" framing collapses into an ordinary backlog item, and the paper's account of why KaNairo lacks interoperability is wrong even if the fact of the lack briefly persists.
3. The NEODC or KNSDI produces a public, Nairobi-specific, building-level dataset within, say, the next 24 months. This would not fully close the gap on its own, but it would falsify the specific claim that Kenya's institutional geospatial infrastructure is "not yet operating at the scale... a building-by-building city twin would need" — the framing in §6 that this paper leans on most heavily.
4. A future, more careful audit finds that Virtual Singapore or Dubai Live, examined at the same level of skepticism applied here to KaNairo, also fails the same maturity thresholds — in which case §7's comparison was unfair not because the frameworks are wrong but because this paper let press coverage stand in for verification on the comparison cases in a way it explicitly refused to do for KaNairo. That would not save KaNairo's placement, but it would damage this paper's argument that the gap is Nairobi-specific rather than universal-and-just-better-marketed elsewhere.
5. One of the three cited maturity frameworks is shown to have been superseded or discredited within the field before this paper's claims are acted on. Least damaging, because two independent frameworks converging on the same placement would likely survive the loss of the third.

10. Limitations

This paper does not claim to have independently re-measured any of KaNairo's underlying figures — the 73.6% height coverage, the 1.27 million building count, the tile architecture — and takes them from the corpus's own prior work rather than re-deriving them. It does not claim comprehensive coverage of every published digital-twin maturity framework; three were chosen for being real, checkable, and independently authored, not because they are the only three that exist or the three the field has settled on. It does not claim to have verified Virtual Singapore's or Dubai Live's own maturity placement with the same rigor applied to KaNairo — §7 says this explicitly and §9's falsification condition 4 names the risk that creates. It does not claim BIM/IFC adoption would be sufficient, on its own, to make KaNairo a twin by any of the three frameworks — it is one necessary piece among several (federation, curation, live sync) that this paper found lacking. It does not make any claim about whether residents, Nairobi City County, or any Kenyan institution should want KaNairo to become a twin in this specific technical sense, or whether the maturity ladder itself represents the right priorities for a city at Nairobi's income level and institutional stage — that is a legitimate, different question this paper does not attempt to answer.

Footnote definitions

Appendix A — Evidence table

S = external source, independently verified for this paper. C = corpus-internal, taken from prior Flux Working Papers and not re-derived. Rows 19–22 are the measurements this paper's argument needs but does not have — they are left empty rather than estimated.

Appendix B — Citation register (CITATION-NEEDED)

1. Whether KaNairo has ever attempted an IFC or CityGML schema export, even experimentally. Not found in the public corpus or in general search. If Flux's engineering team has internal notes on this, it belongs in a future paper, not invented here.
2. Nairobi City County's own institutional stance on KaNairo — whether the county regards it as complementary open data, a dataset it has evaluated and declined to adopt, or has no awareness of it at all. No public statement found either way.
3. The actual proportion of Nairobi building permits submitted with underlying BIM-native design files (even though the county only requires a PDF output). This would materially change §6's claim — if most architecture firms in Nairobi already produce IFC-native models for their own use and simply flatten them to PDF for submission, the sourcing gap is a policy/format-conversion problem, not a data-creation problem, and is considerably easier to close than this paper assumes. No source found; flagged rather than guessed.
4. An independent, rigor-matched audit of Virtual Singapore's and Dubai Live's placement on CITYSTEPS or the Evans/Atkins spectrum. All figures used for both in §7 are press-reported or vendor/government-sourced case studies, not independent academic maturity assessments. This is the single biggest asymmetry in this paper's own method, carried forward from WP37's own Appendix B practice of naming the gap rather than closing it artificially.
5. Whether the Kenya Space Agency's NEODC framework (launched March 2026) has, by the time this paper is read, produced any Nairobi-specific or building-level output. This is explicitly time-sensitive; see falsification condition 3 in §9. A follow-up paper should check this directly rather than let this snapshot stand unexamined.