Smart Cities

A smart city applies sensing, networking and analytics to municipal functions — traffic, water, waste, energy, policing, permitting — and in its stronger form runs the city from an integrated operational picture rather than from departmental silos. The term covers two things that share a name and almost nothing else, and keeping them apart is the whole analytical task. (a) The greenfield smart city: a new settlement designed digitally from the ground up — Songdo, Masdar, Sidewalk Toronto's Quayside, NEOM. (b) Retrofit municipal instrumentation: sensing and integration added to a city that already exists, department by department. Almost all the money and almost all the failures are in (a); almost all the demonstrated benefit is in (b); and evidence from the second is routinely used to sell the first.

This brief is organised around one question, because that question turns out to be the whole subject: has sensing-driven urban management ever produced a measured improvement in a city outcome? The answer, on everything reachable here, is effectively no — not because the instrumentation demonstrably fails, but because the field has almost never run the study. The few genuinely independent evaluations that exist are null or unflattering, the two best-known positive figures come from a party with an interest in them, and the strongest positive result in the largest programme on earth is a transfer result nobody designed for.

The second organising claim is that this is not an accident of a young field. It is a consequence of how these systems are bought. Sensing arrives inside a bundle — a waterfront modernisation, an area-based development, a special-purpose delivery vehicle — and the bundle, not the sensor, is the political and contractual unit. Attribution is destroyed at procurement, before any measurement is attempted. The workback plan in sections 4 and 6 therefore runs through contract clauses and evaluation design, not through better sensors.

Scope boundary and a provenance note. Urban form and density are Sustainable Megacities; housing delivery is Future Housing Systems; the forecasting and accountability machinery this brief leans on is Megaproject Governance. And a disclosure that governs every number below: the research pass behind this edition fetched nothing. Its retrieval channels were closed, and its quantitative spine was recovered from the Institute's own published corpus and citation ledger. Section 14 says exactly what that is worth and what it costs.

Established The flagship greenfield projects have a consistent outcome, and it is not the one the category was sold on. Sidewalk Toronto was cancelled in May 2020, after Alphabet's Sidewalk Labs had spent years developing a proposal for the Quayside site on Toronto's waterfront. Established The project collapsed over data governance — specifically over what Google would do with continuous data about residents' everyday lives. Established It is the most studied failure in the field and the failure was not technical: the project had capital, design talent, political backing and a willing site, and no technical capability was missing. Frontier That is the load-bearing observation of the whole subject, and everything in sections 4, 9 and 10 follows from it.

Established Songdo and Masdar are named in every treatment of this topic, including this one, and this brief publishes no figure for either. Their capital costs, their planned-versus-actual populations, their commercial occupancy, which of their advertised systems were ever commissioned, and whether any party other than the developer has published a measured outcome for them — none of it was obtainable, and none of it exists anywhere in the Institute's corpus. Established The widely circulated cost and population figures for both have a developer or promoter origin and should be attributed rather than asserted; this brief declines to repeat them. Speculative The absence is worth stating plainly because it is characteristic: two projects that anchor the popular case for the entire category have no independently published outcome record that this pass could reach.

Frontier NEOM is the live case, and it is contracting in public. The Line was announced at 170 km long, for 9 million eventual residents, at a headline above $1 trillion — a figure never supported by an approved budget covering the full scope. Frontier In April 2024, reporting placed the 2030 target at 2.4 km built and fewer than 300,000 residents, down from 1.5 million. Frontier In May 2026, reporting indicated work on The Line halted until after 2030 following a strategic review, with the 2030 residency figure at up to 100,000, Trojena and Red Sea assets postponed, and roughly $3 bn redirected to Oxagon for AI data-centre infrastructure. Established Every one of those numbers is a press report of a private programme's internal planning, not an audited figure, and this brief flags them accordingly. Speculative Read against the previous paragraph, the pattern is four flagship attempts — Songdo, Masdar, Quayside, NEOM — and four failures or retreats, across two decades and several political systems. That is a pattern claim, not a tabulation, and the comparative study that would convert it into one has never been produced.

Established By far the largest body of evidence, and the reason this brief can say anything quantitative at all, is India's Smart Cities Mission: 100 cities, about Rs 1.5 lakh crore (US$17.5 billion), roughly 7,555 projects, concluding in 2025 after ten years. Established Nothing else in the field is within an order of magnitude of it for scale, duration or published assessment. Frontier Its results split cleanly, and the split is the most useful finding available anywhere in the subject.

Frontier What worked was the integration layer. Integrated Command and Control Centres absorbed 98.3 per cent of their allocated funds — the highest utilisation of any sector in the programme. Frontier They then proved useful in a way nobody designed them for: during COVID-19 the centres in Ahmedabad, Pune, Bhopal and Varanasi operated as emergency response hubs. Frontier This is the strongest single positive result in the largest programme in the field, and it is worth being precise about what kind of result it is. It is a transfer result — capability built for one purpose proving useful for another — reported by four cities, self-assessed, with no counterfactual and no comparison group. Speculative It is real, it is modest, and it is not an evaluation.

Established What failed was distribution and governance, and here the evidence is measured rather than alleged. Analysis of 60 of the 100 cities found the area-based development model covered about 7 per cent of the average city's area while taking 80 per cent of the funding and serving 9 per cent of the population. Established Delivery ran through Special Purpose Vehicles operating in parallel to elected municipal bodies, which sidelined the constitutional decentralisation that was meant to govern them; in Indore the mandated Smart City Advisory Forum was never constituted at all. Established Transport spending went 40 per cent to roads and parking, 20 per cent to public transport, and 2 per cent to buses. Frontier That last split is a revealed preference rather than an oversight, and it matters for a reason section 3 develops: it means the counterfactual to instrumentation is not hypothetical. The programme itself chose repeatedly between sensing and conventional service provision, and mostly chose roads.

Established The individual project record is specifically and instructively mixed. Chandigarh's 24×7 water pilot delivered muddy water; Surat's cycle tracks went unused for commuting; Bhopal's e-bike scheme declined after its first year; Dharamsala's sensor bins failed physically; and digital waste tracking added workload for sanitation workers without producing systemic improvement. Frontier None of these is an evaluation either. They are reported project outcomes, and they are useful chiefly because they are the failures a programme published about itself.

Established Two outcome figures from the Indian programme circulate as evidence that sensing works, and neither is an evaluated effect. Nagpur reported a 14 per cent decline in overall crime alongside expanded surveillance infrastructure; smart classrooms accompanied a 22 per cent rise in school enrolment across 19 cities. Established Both are associations published by an interested party. Frontier Neither carries a comparison group, a pre-registered outcome, or an evaluator independent of the programme, and the crime figure in particular is a correlation attached to an intervention whose sponsor has an interest in the correlation. Established This brief cites them as what they are.

Established Independent evaluations of sensing-driven municipal interventions do exist — there are very few, and they are not encouraging. The clearest is a peer-reviewed, quasi-experimental evaluation of Chicago's predictive-policing pilot, independent of the police department evaluated: 426 individuals on the Strategic Subject List against 17,754 matched comparisons, finding no difference in the likelihood of becoming a homicide or shooting victim, and finding the apparent city-level decline to be part of a pre-existing trend. Established A 2026 audit of urban AI systems examined 28 implementations selected from 157 deployments across six domains between 2015 and 2024, and reports the pairing that ought to govern how this entire field is read: ShotSpotter, an acoustic gunshot-detection sensor network, at 97 per cent acoustic accuracy yielding 9.1 per cent crime-fighting effectiveness; Chicago's predictive-policing list with 16.3 per cent of listed individuals confirmed as gang members against commanders' belief of 95 per cent; COMPAS at roughly 60 to 70 per cent overall accuracy with a 44.9 per cent false-positive rate for Black defendants against 23.5 per cent for white defendants. Frontier The audit's argument is that high metric accuracy is not merely compatible with policy failure but is sometimes the mechanism of it — the instrument's own performance number substitutes for the outcome it was bought to move.

Established The single highest-standing design anywhere in citizen-sensing urban management is a randomised trial, and it found nothing. Across 200 zones in an African capital, 50 citizen reporters in each of 100 treated neighbourhoods generated 23,856 reports over nine months, measured against 679 physically inspected waste piles. Treatment effects on waste-pile size were 4.23 m² smaller at five months (p = .112) and 7.78 m² smaller at nine months (p = .303). Established Neither was significant. There was no significant increase in cleanups; a modest reduction in burning did not persist; citizen response ran around 10 per cent; and the city eventually abandoned the programme over cost and report reliability. Frontier That is what a properly powered, independently designed test of “instrument the city and act on what it tells you” looks like, and the field has approximately one of them.

Established The one large municipal sensing dataset with a real denominator points the causal arrow the other way. A peer-reviewed analysis of a complete street-reporting platform population — 399,364 reports from 154,957 unique users over six years — found a fix rate of 39.9 per cent, and that users whose first report was fixed were 57 per cent more likely to file a second, at 24.1 per cent against 13.6 per cent. Established Read carefully, that establishes that government responsiveness drives citizen participation. It does not establish that the platform improved street repair: 39.9 per cent is a fix rate with no counterfactual. Frontier The direction — state behaviour causing citizen behaviour — is the reverse of what the advocacy literature usually claims.

Established And the comparison that should discipline the whole subject involves no technology at all. A panel analysis across 3,651 comparable areas covering Brazil's municipalities from 1990 to 2004 found that participatory budgeting raised the health-and-sanitation budget share by 2 to 3 percentage points — 20 to 30 per cent of that category's baseline share — and reduced infant mortality by 1 to 2 per 1,000, about 5 to 10 per cent of the baseline rate. Established That was face-to-face, offline, and in the 1990s. It changed municipal spending and it changed a health outcome, with an identification strategy and a comparison group. Frontier Nothing in the instrumented-city literature approaches it for evidential strength, and the gap is not a gap in the technology.

Established The bottom line of this section is a negative and it is the brief's spine. Almost nothing in this field has been evaluated as an intervention. The strongest available positive results are one transfer result without a counterfactual and two associations published by an interested party; the strongest independent results are null or adverse; and the programme's own ministry initiated impact assessment only after missing its delivery deadlines, which is the wrong order and is the structural reason the field's central question is still open.

Frontier Is the integration layer the actual product? The command-and-control centres outperformed everything else in the Indian programme, including their own design intent, while the sensor estates beneath them produced the failure list in section 2. If that generalises, the useful smart city is a data-integration and situational-awareness project inside municipal government, and the sensors are largely incidental — a far smaller, far less saleable proposition than the one marketed. Speculative The experiment that would settle it is specific and cheap and nobody has run it: fund the command centre without the sensor programme in a matched set of cities, and compare.

Frontier Why has every greenfield attempt failed or shrunk? Two accounts are live and they predict different things. On the first, cities are emergent systems whose behaviour depends on occupants who do not yet exist and cannot be specified in advance, so designing one as a technical system is a category error. On the second, these were real-estate ventures carrying a technology narrative, and they failed as real estate for ordinary real-estate reasons. Speculative The record does not separate them, and separating them requires exactly the per-project financial and occupancy tabulation that sections 2 and 14 record as missing.

Frontier Can the surveillance-dependent benefits be obtained without the surveillance? The weak form of the objection says these systems collect more than they need and keep it too long, and that governance fixes it. Speculative The strong form says the value proposition requires individuation: a system that optimises flows must distinguish the entities being optimised, and aggregate data cannot support the personalised services used to justify the spend. On that reading, the privacy-preserving smart city is not a constrained version of the product but a different and much less valuable one — which would explain why it is repeatedly proposed and then quietly descoped. Frontier The decisive test is runnable and unrun: measure benefit magnitude under aggregate-only processing against individuated processing on the same deployment. If the aggregate-only version performs comparably, the strong form fails.

Frontier Did the sensing change a decision? The research question in this field is shifting from “can we measure it” to decision linkage — tracing a sample of sensor-derived alerts through to documented actions and outcomes. Established Cities already hold far more data than they act on; a sensor that reports a pothole to a department with no repair budget changes nothing. Speculative A decision-linkage audit needs a records request rather than a deployment, which makes it the cheapest high-value unpublished study in the subject.

Frontier Urban digital twins. Whole-city twins are technically achievable and their operational value beyond specific engineering uses remains largely asserted rather than demonstrated. Speculative The unresolved question is what a twin is validated against: a model calibrated on the same telemetry it is meant to interrogate has no independent standard of correctness. Handwave A national-academies report on foundational research gaps for digital twins sits in this project's citation backbone marked unconfirmed and has never been read in any pass, so this brief states no finding from it.

Frontier Does urban experimentation accumulate? A 2026 synthesis drew on eight separate databases covering close to two thousand urban experiments and distilled ten lessons across process, politics and impact. Frontier Two thousand experiments yielding ten lessons is an enormous denominator for a small output, and the existence of eight databases rather than one is itself evidence that outcomes are not measured commensurably enough to accumulate. Speculative On this reading the field has produced knowledge without producing capacity, which is a different failure from the one its critics usually allege.

Handwave And the question nobody is positioned to ask. Does continuous municipal instrumentation change what a city optimises for, in ways nobody chose? A city that measures traffic flow well and social cohesion not at all will improve traffic flow. Handwave There is no evidence either way, the claim is not currently falsifiable as stated, and it is nonetheless the most interesting hypothesis in the subject. Speculative The testable version, and the most useful general claim the evidence suggests, is narrower: instrumentation succeeds where it gives an existing authority a better view of something it already does, and fails where it is expected to substitute for a service that is not there. That would predict that the success-failure split in section 2 tracks pre-existing institutional capacity rather than technology type. It has not been tested.

Established The binding bottleneck is data governance, and Sidewalk Toronto is the proof. A project with capital, design talent, political backing and a willing site died on the question of who owns and controls the data. Any programme treating that as a compliance annex rather than as the core design problem is re-running the experiment. Frontier The governance prescription the post-mortems converge on is three-part and unglamorous: published rules on collection, retention, anonymity, commercialisation and ownership, in force before tender; municipal expertise sufficient to hold them; and transparency obligations that go beyond informed consent, because consent has no meaningful form in public space.

Established Second, municipal institutional capacity. Cities are asked to specify, procure and supervise systems whose vendors understand them far better than the buyer does, which produces lock-in and unusable procurement. Established The Indian programme's answer — route delivery through Special Purpose Vehicles outside the elected structure — solved capacity by bypassing accountability, and the published assessment is clear that this was a cost rather than a fix.

Frontier Third, and least discussed: operating cost and lifecycle. Sensors, networks and analytics platforms are capital projects carrying permanent operating obligations, funded from capital budgets, in institutions whose revenue budgets are already strained. Dharamsala's failed sensor bins are the small version of a general problem: the grant builds it, and nothing funds its twelfth year. Speculative The strong version of this claim needs published survival curves for municipal sensor estates and an operating cost per sensor-year. Established No such figures exist in any pass of this project, and none were found in the corpus. Deployment counts are therefore a stock reported as a flow.

Frontier Fourth, integration semantics. The genuinely hard engineering is making a water department's asset identifier mean the same thing as the roads department's. That is where these programmes stall, and it is unglamorous enough that it is rarely funded as the primary work. Handwave Proprietary platforms procured per department reproduce, in software, exactly the siloing the programme was meant to end.

Frontier Fifth, and the one that binds everything else: attribution is destroyed at procurement. Sensing is almost never deployed alone. It arrives with new signals, resurfaced roads, a reorganised department, and a delivery vehicle that reports to nobody the electorate chose. When the bundle is the contractual unit, no subsequent measurement can separate the sensing from the works, and no independent evaluator can obtain the data anyway, because the data sit with the vendor. Established That is why the workback plan in section 6 runs through contract clauses.

Frontier Progress depends first on municipal data governance settled before procurement rather than negotiated under it: published rules on collection, retention, anonymity, commercialisation and ownership, which is precisely the list the Sidewalk post-mortems converge on. Frontier Second, in-house technical capacity sufficient to be an informed buyer, measurable against vendor lock-in incidence rather than against headcount. Frontier Third, open standards and data portability as procurement conditions, testable by a concrete acceptance criterion: a data export completed by a second vendor without the incumbent's cooperation. Frontier Fourth, lifecycle funding for operations rather than capital-only grants, since a capital grant with no revenue line is a decision to abandon the estate on a schedule. Speculative Fifth, and least often named, an evaluation right reserved in the contract — a named independent evaluator, guaranteed data access, and a pre-specified outcome metric whose definition cannot be changed alongside the deployment.

Frontier Two dependencies are structural rather than contractual and are usually left out of the list. The first is the shape of the vendor market: consolidation among smart-city suppliers determines how credible a portability clause is, because a clause is only as good as the existence of a second supplier able to honour it. Speculative Nobody has published the concentration figures for municipal platform procurement, and until somebody does, “avoid lock-in” is advice without a denominator. Frontier The second is a settled operating-cost basis. Until a city can state a cost per sensor-year and an expected estate life, it cannot price the twelve-year obligation it is signing, and the capital-grant model will keep producing estates that decay quietly between funding cycles. Established Neither of these is a research dependency in the scientific sense. Both are facts about the world that nobody has bothered to measure.

Established On the map, this brief depends directly on Megaproject Governance for the forecasting and accountability structure, because the smart-city record is largely a governance record and the base rates that apply to any large programme apply here. Frontier It connects to Sustainable Megacities on urban form and on the measurement problem underneath both, to AI Governance and AI-Assisted Governance where municipal analytics becomes municipal decision-making, to Civic Technology for the participation side and for the evidential standard this brief borrows from it, and to Infrastructure Resilience for the failure modes an instrumented city acquires along with its instruments.

Established The cheapest decisive study in the subject already has its comparison group and has not been run. India's Mission gave 100 cities staggered implementation across ten years, and it shortlisted cities that were not ultimately selected. That is a difference-in-differences design sitting in public administrative data: treated cities against shortlisted-but-not-selected controls, on outcome metrics the cities were reporting before the programme began. Frontier It requires no deployment, no new instrumentation and no vendor cooperation. It requires someone to fund an analyst. Speculative Its absence, after US$17.5 billion, is the single most informative fact in this brief.

Frontier Second, one evaluated intervention, done properly. Randomised or staggered rollout of a single, bounded intervention — adaptive signal control, leak detection, transit priority — with a pre-specified city-level outcome, an effect size and a confidence interval. Established This is entirely feasible at city scale; the waste-reporting trial in section 2 demonstrates that a municipal randomised design can be fielded and analysed. Frontier The design must satisfy five conditions the field routinely violates: the metric is defined before deployment and its definition does not change across the deployment boundary; the baseline is long enough to characterise variance; there is a control or synthetic-control area; the sensing is not co-deployed with conventional works, or the works are separately identified; and the evaluator is independent of both vendor and procuring department.

Frontier Third, the integration-layer test. Fund command centres without sensor programmes in a matched set of cities, and compare. If situational awareness is the product, this is the experiment that shows it, and it would substantially reprice the market.

Frontier Fourth, the aggregate-only comparison. Run the same deployment under aggregate-only processing and under individuated processing, and measure the benefit difference. This is the only test that adjudicates the strong form of the surveillance objection, and it is cheap where a system is already running.

Speculative Fifth, the decision-linkage audit — trace a random sample of sensor-derived alerts to documented actions and measured outcomes — and sixth, a sensor-estate survival study: operating cost per sensor-year and estate survival at year twelve, across a cohort of deployments with known install dates. Frontier Both are records exercises. Established Neither has been published.

Frontier And a specific re-evaluation. Nagpur's 14 per cent crime figure needs a comparison against matched cities before it is cited as a result rather than as a correlation attached to a programme with an interest in it.

Established The engineering requirements are conventional systems engineering, and the difficulty is scale and longevity rather than novelty. Sensor networks with power and communications budgets that survive a decade outdoors in weather, dust and vandalism; municipal data platforms with defined schemas across departments; identity and access control adequate to hold whatever governance rules the city has published; and network coverage over territory the city does not fully control. Frontier Nothing on that list is a research problem. Every item on it is a procurement and maintenance problem.

Frontier The genuinely hard part is semantic rather than physical. Reconciling asset identifiers, coordinate conventions, update cadences and lifecycle states across water, roads, waste and planning is where integration projects actually stall, and the fraction of identifiers reconciled across departments is a measurable and almost never reported figure. Speculative A useful acceptance test for any municipal platform: can a second supplier read the estate without the incumbent's help?

Frontier Sensor fusion and computer vision are quietly replacing dedicated sensors, which changes the cost profile and the privacy profile at the same time — a camera that counts vehicles is also a camera. Frontier Edge processing has been adopted specifically as a privacy-preserving architecture: process locally, transmit aggregates. Established It changes who holds the data. It does not change whether the city watches, and it does not answer the aggregate-only benefit question in section 3.

Frontier The connectivity question has quietly become a public-infrastructure question. Low-power wide-area municipal radio — LoRaWAN and its equivalents — is cheap enough that a city can own its own network rather than rent capacity from a carrier, which changes who can switch a deployment off and who sees the traffic. Speculative Whether municipally owned radio actually reduces lock-in, or merely relocates it into the platform layer above, is unmeasured. Established The durable engineering requirement underneath all of it is boring and rarely specified: an estate register recording what was installed, when, by whom, under what warranty, and whether it still reports — the absence of which is why nobody can produce a survival curve.

Handwave Whole-city digital twins are achievable as engineering and unvalidated as operational instruments, for the reason given above: there is no independent standard against which to check a model built on the same feeds it is meant to interrogate. Speculative A twin that could be falsified would need an outcome it predicts and that somebody measures independently, which returns the question to section 6.

Established The adjacent fields that matter here are mostly not technical. Municipal public administration and public finance determine whether an operating obligation can be carried at all. Public procurement law and contract theory are where data ownership, exit rights and evaluation rights are either reserved or lost, and therefore where the measurability of this entire field is decided. Privacy and surveillance law set what may be collected and retained, and increasingly what must be disclosed. Transport modelling and operations supply most of the concrete use cases with short causal chains. Sensor networks, low-power wide-area radio and edge computing are the physical substrate; municipal algorithmic accountability and automated-decision registers are the emerging disclosure layer.

Frontier Two adjacencies are less obvious and more load-bearing. Programme evaluation methodology — difference-in-differences, synthetic control, interrupted time series, stepped-wedge designs — is the discipline this field has conspicuously failed to import, despite the tools being standard, cheap and well understood elsewhere in public policy. Frontier And the participation literature, because it is the one municipal domain with a comparably sized evidence base and it has run the studies this one has not.

Frontier Surveillance studies and platform urbanism supply the critical vocabulary, and they are adjacent rather than peripheral for a concrete reason: predictive policing is the most studied and most criticised municipal sensing application, and it is also the one with the strongest independent evaluation evidence, so the critical literature and the quantitative literature converge on the same case. Speculative Anyone assessing a proposed deployment gets more from that convergence than from either side alone.

Established On this map, see Sustainable Megacities for urban form and the urban measurement problem, Megaproject Governance for why large programmes miss their forecasts, AI-Assisted Governance for automated municipal decision-making and its litigated record, Civic Technology for the participation platforms and their denominators, Robotics in Infrastructure for automated inspection, and Infrastructure Resilience for what an instrumented city acquires along with its instruments.

Established This is the topic's centre of gravity, and the Indian record states it plainly. The programme's most consequential design choice was institutional, not technical: delivery through Special Purpose Vehicles running parallel to elected municipal bodies. It moved quickly, and it detached the spending from the accountability structure that India's constitutional decentralisation had established, with a mandated advisory forum in at least one city never convened at all. Frontier The general mechanism, which is what makes this more than a single-country finding: when delivery is bundled into a vehicle outside the elected structure, attribution is destroyed before measurement can begin. Outcome data stop being traceable to spend, because the entity that spent is not the entity that reports.

Frontier The distributional finding is an institutional choice too. Seven per cent of area, 80 per cent of funding, 9 per cent of population is what a programme judged on demonstrations will produce, because a showcase district can be photographed and a citywide water network cannot. Handwave The uncomfortable synthesis is that the institutional failure mode here is legibility rather than incompetence: these programmes optimise for what can be shown, and they do it competently.

Frontier The disclosure layer is thin and knows it. A peer-reviewed analysis of the Amsterdam and Helsinki municipal algorithm registers, conducted when they held three and five entries respectively, found the entries scoped to uncontentious municipal uses and omitting the sectors actually implicated in algorithmic discrimination. Established The counts are now stale — Helsinki has nine, and Amsterdam's systems have moved into the Dutch national register — and the critique survives the growth, because it was never about the count. Frontier Mandatory regimes are appearing: the United Kingdom's Algorithmic Transparency Recording Standard was made mandatory across government departments in 2025. Speculative Whether a mandatory register changes procurement behaviour, as opposed to producing a compliance artefact, is unmeasured and is a tractable study: compare contract terms before and after a mandate.

Frontier The institutional ask is therefore short and expensive to grant. Data governance in force before tender; an evaluation right and a named independent evaluator in the contract; portability tested rather than promised; a revenue line for operations; and delivery inside the accountable structure rather than beside it. Established Every one of those is a decision a council can take without any new technology, and the reason they are rarely taken is that each one makes a proposal look dearer and slower at the moment approval is sought — the same approval-stage pathology documented at length in Megaproject Governance.

Established Surveillance is the unavoidable centre, and it is a design objection rather than a side-effect objection. A city that senses continuously is a city that watches. The benefits accrue diffusely and the costs fall hardest on people already policed most — Nagpur's crime decline and its expanded surveillance infrastructure are the same sentence, and nothing in the published record separates them. Established Consent has no meaningful form in public space, which is why the Sidewalk post-mortems land on published rules and institutional capacity rather than on informed consent. Frontier And the strongest evidence that this is a design objection is the case itself: a project with every technical and financial advantage died on this alone.

Established Error is distributed as unequally as attention. The audited figures in section 2 include a 44.9 per cent false-positive rate for Black defendants against 23.5 per cent for white defendants on one widely deployed risk instrument, and a gunshot-detection network whose 97 per cent acoustic accuracy converted into 9.1 per cent crime-fighting effectiveness. Frontier A system can be accurate about the thing it measures and wrong about the thing it is used for, and the people who bear that gap are not the people who bought it.

Established Distribution is the second issue and it is measured: 80 per cent of funding to 7 per cent of area serving 9 per cent of population is a regressive allocation inside a programme framed as urban improvement. Frontier Displacement compounds it — Indore residents were displaced without consultation, many classed as encroachers and denied compensation, with land monetisation sometimes bypassing existing compensation frameworks.

Frontier And there is a labour dimension the technology framing obscures entirely. Digital waste tracking added workload for sanitation workers whose safety conditions did not improve. Speculative Instrumentation makes work legible to whoever holds the console before it makes it safer, and the legibility arrives first as measurement of the worker. Handwave The general form of this — that the console holder's account becomes the authoritative one and residents' accounts become anecdote — holds even under perfect good faith, and almost no deployment provides an appeal route against a sensor-derived determination.

Established Most humans now live in cities, so municipal operational competence is a civilizational variable whether or not it is exciting. This brief prints no urbanisation percentage, because the figures in circulation mix definitions of “urban” and the primary sources were not obtainable — see Sustainable Megacities, which refuses them for the same reason. Frontier The direction is not in doubt and the precision is.

Established What the evidence supports is modest and worth having. Integration of municipal data improves situational awareness, and that situational awareness proved transferable to an emergency nobody had planned for. Frontier That is a real capability, it is cheaper than the sensor programmes sold alongside it, and it is a defensible reason to keep building integration layers while the evaluation question stays open.

Handwave What the evidence does not support is the framing that gave the category its name — that a city can be designed as a technical system and thereby made to work. Four flagship attempts, four failures or retreats, across two decades and several political systems, with a comparative study still unwritten.

Speculative The genuine long-run question is not whether cities become instrumented. They will, incrementally and unevenly, in the way municipal water telemetry already did. It is whether the instrumentation is governed by the people measured or by whoever supplied the sensors — and that is being decided now, in procurement documents, by institutions that mostly lack the capacity to read them. Handwave Cities outlive every technology installed in them, which is the strongest available argument for portability over capability.

Speculative There is a further implication that the field's own framing hides. If the useful object turns out to be the integration layer rather than the sensor estate, then the civilizational upgrade available here is administrative: a municipal government that can see itself whole. Handwave That is a smaller and much older ambition than the one the category advertised, it costs a fraction of what has been spent pursuing the larger one, and it is the version of this subject that is most likely to survive the marketing era that named it.

The technology is available now and has been for a decade, which makes this an unusual timeline section: nothing below waits on an invention. These horizons track municipal capacity, procurement practice, governance settlement and operating budgets, and the single variable that moves all of them is whether anybody funds an evaluation. A reader should treat the dates as forecasts about institutions, not about engineering.

• 10 yr: Frontier Integration layers and command centres normalised in large cities, on the strength of the Indian evidence rather than on any evaluation of it; greenfield announcements continue and continue to shrink; data governance frameworks still lagging procurement in most jurisdictions. Speculative The realistic best case is that two or three properly designed evaluations exist by then, and the difference-in-differences re-analysis of the Indian Mission is much the likeliest of them, because it needs an analyst and administrative data rather than fieldwork. Frontier Mandatory algorithmic disclosure spreads faster than evaluation does, because it is cheaper and more visible.
• 25 yr: Frontier Enough evaluated intervention data to know which municipal instrumentation earns its operating cost, assuming the studies above get funded at all. Speculative Expect a much shorter list than is currently marketed, and expect leak detection and adaptive signal control to be on it, because their causal chains are short, their counterfactuals are cheap and their outcomes are already metered by an existing utility. Speculative Expect the surveillance-dependent applications to be adjudicated politically rather than empirically, because the aggregate-only comparison will still not have been run.
• 50 yr: Speculative Instrumentation ordinary and invisible, in the way municipal water telemetry already is; the term “smart city” probably obsolete, having named a marketing era rather than a technology. Speculative The interesting residue is institutional rather than technical: whether an evaluation right and tested data portability became standard contract terms. Frontier That is decidable now, at negligible cost, and it is what would be visible fifty years out.
• 100 / 250+ yr: Handwave No basis exists for forecasting urban governance arrangements on this horizon, and a brief that offered one would be doing something other than assessment. Handwave The one durable datum is that cities outlive every technology installed in them, which argues for instruments that can be removed and replaced without removing the institution that reads them.

• Depends on Municipal data governance settled and published before tender rather than negotiated under it; in-house technical capacity sufficient to specify, procure and supervise systems whose vendors know them better than the buyer does; open standards and tested data portability as procurement conditions rather than aspirations, with the test being an export completed by a second supplier without the incumbent's help; lifecycle funding for operations rather than capital-only grants, priced against an operating cost per sensor-year that nobody currently publishes; a reserved evaluation right with a named independent evaluator, guaranteed data access and an outcome metric whose definition cannot change alongside the deployment; and delivery inside the accountable municipal structure rather than through a parallel vehicle that detaches spend from the body the electorate chose. Every one of these is a governance and contracting decision available today. None requires a technical advance, and the reason they are rarely taken is that each makes a proposal look dearer and slower at the moment approval is sought. The dependency this brief carries formally is Megaproject Governance, because the base rates and approval-stage incentives documented there govern any programme of this size and are the reason the pathology repeats across political systems.
• Enables A municipal situational-awareness capability that transfers to emergencies nobody planned for, which is the one benefit the record actually supports and the cheapest thing on this page. Evaluated, targeted service improvements with published effect sizes and confidence intervals, which the field does not currently possess for any intervention anywhere. A governed rather than vendor-defined urban data layer, and with it the ability to answer the question this brief cannot: whether instrumentation improves outcomes, and by how much, for whom, and for how long after the funding stops. Downstream of that answer sit the appraisal assumptions of every other infrastructure slot on this map that expects to be operated on its own telemetry — corridors, ports, grids, transit and water systems are all sold partly on the promise that instrumenting them will improve how they run, and that promise currently rests on an evidence base this brief has just described.
• Adjacent Municipal public administration and public finance, which decide whether an operating obligation can be carried at all; public procurement law and contract theory, which is where the measurability of the whole field is actually decided; privacy and surveillance law, including the emerging disclosure regimes for automated municipal decisions; transport modelling and operations, which supply most of the use cases with short causal chains; sensor networks, low-power wide-area radio and edge computing as the physical substrate; sensor fusion and computer vision, which are replacing dedicated sensors and shifting the cost and privacy profiles at once; surveillance studies and platform urbanism for the critical vocabulary; and — the adjacency this field has most conspicuously failed to import — programme evaluation methodology, whose standard quasi-experimental tools are cheap, decades old, well understood, and almost entirely unused here.

Established “The smart-city failures were technical.” Sidewalk Toronto had the capital, the design talent, the political backing and the site, and it died on data governance. Established Treating that as an engineering problem misreads the single most studied case in the field, and it is the error that most reliably produces a repeat.

Frontier “Smart city names one thing.” It conflates greenfield construction, which has a uniform record of failure and retreat, with retrofit instrumentation, which has a mixed record containing real successes. Established They share no common technology, actor, risk profile or evidence base. Frontier If the category is incoherent, then both “smart cities work” and “smart cities failed” are close to unfalsifiable, and the correct move is to dissolve the category and evaluate each intervention on its own terms. Established The practical consequence is a marketing one: evidence from retrofit is routinely used to sell greenfield.

Handwave “A showcase district is a smart city.” India's area-based model reached about 7 per cent of the average city's area with 80 per cent of the money, serving 9 per cent of the people — and it is the part that appears in the photographs.

Established “Reported outcomes are measured effects.” The 14 per cent crime decline and the 22 per cent enrolment rise are associations published by an interested party, not causal estimates. Frontier The general form of the error is worth naming, because it is how most benefit claims in this field are manufactured: a metric is redefined alongside the deployment, so the before and after are not the same quantity. Speculative Continuity of a metric's definition across the deployment boundary is the sharpest single test a reader can apply to any smart-city claim.

Established “The sensors are the expensive part.” The capital grant installs them; nothing funds their twelfth year. Dharamsala's failed sensor bins are a more representative outcome than the launch photography, and deployment counts are a stock reported as a flow.

Frontier Enthusiast-side: “India's Smart Cities Mission proves the concept works.” It is simultaneously the largest evidence base in the field and its most damning distributional finding, and both readings come from the same programme. Established A 7-per-cent-of-area, 80-per-cent-of-funding allocation and a 40 / 20 / 2 transport split between roads, public transport and buses are not compatible with a simple success narrative.

Frontier Enthusiast-side: “The command centres validate the sensor programme.” They validate the integration layer. If the integration-layer hypothesis holds, the sensors are largely incidental — which is a far smaller and less saleable proposition than the one that was marketed, and it is the reason the test in section 6 is not run.

Speculative Enthusiast-side: “Privacy-preserving architecture makes the surveillance objection go away.” Edge processing changes who holds the data, not whether the city watches. Frontier Asserting that privacy and functionality are always reconcilable, without running the aggregate-only benefit comparison, is assuming the conclusion of the one experiment that would settle it.

Handwave Enthusiast-side: “Digital twins are ready for operational city management.” Technically achievable; operationally asserted. The validation question is unanswered and the national-academies review that might bear on it has not been read in any pass of this project.

Frontier Sceptic-side, for balance: “None of it works.” Not supported. The command-centre utilisation figure and the COVID-19 transfer are real. Established The honest statement is that what the evidence supports is modest and worth having — integration of municipal data improves situational awareness, and that proved transferable to an emergency nobody planned for. Frontier Sceptic-side error two: “the flagships were abandoned, therefore the idea failed.” Abandonment of the greenfield model is entirely compatible with retrofit instrumentation succeeding, and the flagship post-mortems should not be allowed to do the work of evaluating the whole field.

Established And the misconception this brief is most at risk of creating: “absence of evidence is evidence of absence.” It is not, in general. Frontier But the inference is stronger than usual here, because the studies are cheap, the tools are standard, the comparison groups exist, and the expenditure has been enormous. Frontier For an intervention class this well funded and this long-running, twenty years of non-evaluation is itself a finding about the field's incentives rather than a neutral gap in the literature. Speculative The honest terminal position is not “sensing does not improve city outcomes.” It is that the question has been made unanswerable by how these systems are bought — and that is a different claim, with a different remedy, and the remedy is a contract clause.

Established Finally, a disclosure about this page rather than about its subject. The research pass that produced this edition fetched nothing at all: its bibliographic search channel was returning a frozen, off-topic payload, and publisher and agency hosts refused every request. The quantitative spine above — the Mission's scale, the 98.3 per cent utilisation figure, the 7 / 80 / 9 split, the project outcomes, the Chicago evaluation, the waste-reporting trial, the NEOM sequence — was recovered from the Institute's own published briefs and from its citation and URL ledgers, where an earlier pass with working retrieval had read and recorded them. Frontier Those claims are labelled internally as read by a prior pass, not re-read in this one; where a figure's source URL carries a recorded live check, this brief's reading list says so. Established That recovery is a genuine property of a corpus that keeps its own citation ledger: it can survive a retrieval outage that would otherwise have produced a page of hedged nothing. Frontier It is also the exact property that would let an error persist unexamined across releases. A number that entered the corpus wrong would be recovered, re-published and re-cited with the same fluency as a number that entered it right, and the ledger records only that a URL responded, never that the sentence attached to it is true. Speculative The corrective is unglamorous and belongs in this section because it is a claim about reliability: a corpus that recovers from itself needs a scheduled re-fetch of its own load-bearing citations, not merely a link check — and this brief's own highest-priority re-fetch is the Nature assessment that carries most of the numbers above.