A Reading Is Not a Decision

Every smart-city procurement ends with a number appearing on a screen. Almost none of them specify what is supposed to happen next. That gap is where this atlas has spent two years accumulating evidence, and the pattern that has emerged is not the one the sector talks about. The hard part is never the sensing. It is whether anyone was holding the legal power to act on what the sensor found. Cities buy the measurement and assume the consequence. The consequence turns out to be a separate purchase, and usually a legal one rather than a technical one.

Six rungs between a reading and an act

Sort the atlas's monitoring entries not by technology or by city but by what the number is permitted to do, and they fall onto a ladder with six rungs.

1. Nothing follows. The reading is published. No rule attaches to it.
2. A warning reaches the public. Singapore's drainage network texts residents at 50, 75, 90 and 100 per cent of canal depth. Copenhagen's harbour baths turn red after a sewer overflow. The citizen is now informed and must act alone.
3. Staff decide, with the data in front of them. Bangkok's flood control centre forecasts street-level depth at four-metre resolution; crews still work the 200 pumping stations by hand.
4. The threshold has legal force, but a person validates it first. Bogotá writes its index bands into a joint resolution, then a standing panel confirms the reading before any restriction bites.
5. The threshold has legal force automatically. Delhi and Mexico City. A number crosses a line and construction stops.
6. The observation is the act. No decision sits in between at all.

Almost every disappointment in the corpus is a city that bought rung one and expected rung five.

Rung zero: a court can compel measurement, not action

In 2019 thirty-two residents sued Indonesia's president, three ministers and three governors over the air in Jakarta. On 16 September 2021 the Central Jakarta District Court found all seven had acted unlawfully; the appeal failed and the Supreme Court made the judgment final in November 2023. The city's monitoring network grew from five reference stations to 111 points feeding a public portal.

And then nothing attaches. No index threshold in Jakarta carries the automatic legal consequence Delhi's does. The emission-test fines that did bite were suspended. Residents won a lawsuit, and what they won was better information about the air they are still breathing.

Warning outran actuation

The two hardest cases in the atlas sit on rung two, and both involve underpasses.

In July 2020 three drivers drowned in Busan's Choryang underpass. The city responded properly by the standards of the sector: it mapped inundation risk across 201 neighbourhoods and put the result online in 2023, merging flood sensors, rain gauges, river gauges and tide readings into a portal that grades every area safe, caution, alert or danger. Four years after Choryang, 41 of the city's 57 underpasses still had no automatic entry barrier.

Busan can tell you an underpass is filling before it can close it.

Medellín's SIATA runs roughly 1,200 sensors, radars and cameras across ten municipalities, operated by a university under contract with the metropolitan authority since 2010. It has prompted real evacuations. In January 2023 its cameras captured a fatal flash flood in real time at an underpass that was never closed.

Neither city has a sensing problem. Both have a barrier problem, which is to say a capital-works and legal-authority problem wearing a technology costume.

Detected, and then what

Jakarta ETLE: notices police could actually post police statement · budget-limited ~7 %

Bangkok low-emission-zone pilot: fines issued city-reported · Jan 2025 pilot 1.8 %

Jakarta ETLE: violations ultimately confirmed police statement · traffic director 0.01 %

One measure, three points on the same funnel. Jakarta's two rows are stages of one process, not rival figures: police could post roughly 800 of some 12,000 daily camera-detected violations because the postage budget was limited, and confirmation falls further still. Bangkok's figure covers a two-day pilot inside the Ratchadaphisek ring road. All three are police or city statements, not independent audits; none has been checked against case records.

The atlas entry for Jakarta's camera network puts it in one line: automating detection is the easy half, and a camera network is only as strong as the address database and the postage budget behind it. That is not a metaphor. It is a line item.

The top of the ladder, and why almost nobody is on it

About 21 square kilometres of western Osaka lie below sea level, shielded by seven tide gates on five rivers. Since 2018 the prefecture's flood control office has closed them automatically on the national alert signal: a Japan Meteorological Agency tsunami advisory sends a close command to every gate. The three arch gates built in 1970 joined in April 2023, sounding an eight-minute warning to boats first.

This is rung six, and it is worth asking why Osaka could get there when Busan could not. The answer is not budget or engineering culture. It is that the decision contains no judgement. A tsunami advisory is binary, it is issued by a national authority with unambiguous jurisdiction, the time budget is minutes, and no reasonable official would decide otherwise. Automation is available exactly where discretion is not needed.

Where discretion is needed, the ladder gets slippery. Bogotá's index bands sit in a formal resolution, yet a validation panel confirms the reading before restrictions apply. That is not bureaucratic drag for its own sake: air readings can be wrong, and a false alarm that halts a city has costs of its own. But it means the atlas entry's own conclusion holds, that a threshold can carry legal force without carrying automatic force.

Delhi tried to automate a judgement call. Its Graded Response Action Plan has made an air-quality reading legally binding since 2017: defined thresholds trigger construction bans, diesel-generator bans, freight restrictions and school closures across the capital region. A 2022 revision even shifted the trigger to forecast values so stages could be invoked before the air worsened. In 2025 the stages were still invoked only after the index had already crossed the line on 13 of 17 occasions.

A perfect trigger on the wrong lever

Here is the finding that should unsettle anyone who reads the ladder as a maturity model to climb.

Mexico City has the oldest measurement-with-consequence mechanism in the atlas. Manual sampling began in 1966, the automatic network in 1986. Since 2016 a reading of 155 parts per billion of ozone declares a Phase I contingency, and the declaration binds: the plate-based driving ban doubles, factories must cut ozone precursors by 40 per cent, building sites over 5,000 square metres stop, and the Tula refinery is capped at 75 per cent of capacity. The threshold has never been loosened. Peer-reviewed evaluation found it was tightened in 2016, not relaxed.

And the headline lever it reaches for does not work. Peer-reviewed analysis of the Saturday extension found no improvement in air quality, consistent with the earlier finding that households respond to plate rationing by buying a second, older car. The trigger fires exactly as designed, on time, with legal force, into a policy that the evidence says does not clean the air.

A well-built trigger attached to an ineffective instrument produces confident, punctual, measurable futility. That is a worse failure mode than a warning nobody acts on, because it looks like success from the control room.

It was never a sensing problem

Bangkok makes the real constraint unmistakable, because the city's powers run out mid-sentence. When district readings go red, Bangkok closes its own schools outright: 352 of 437 on 24 January 2025. To keep lorries out of nine inner districts it has to borrow powers from disaster law. To get office workers to stay home it can only ask, because work-from-home is voluntary. And a peer-reviewed source-attribution model puts roughly 30 per cent of the region's haze-season fine particulates on burning outside Thailand, where the governor has no standing at all. One threshold, four different levels of authority, four different outcomes.

If that reads like a developing-country constraint, consider Miami-Dade County. Its heat office got the National Weather Service to lower the advisory threshold from a heat index of 108°F to 105°F, which is a textbook rung-four achievement: a measured value, newly consequential. Florida then barred counties from requiring heat protection for outdoor workers, and the county cut the officer post against a budget gap. The measurement improved and the authority to attach anything to it was removed, in that order.

The catch

• The ladder is descriptive, not a maturity model. Rung six suits binary hazards with minute-long time budgets. Applied to a judgement call it produces automated error, and a human validator is then a feature rather than friction.
• Almost every enforcement figure here is self-reported. Bangkok's fine count, Jakarta's confirmation rate and Beijing's trip totals are police or agency statements. Where independent evaluation exists it usually revises the claim downward.
• Absence of a trigger is not always failure. A city may have decided, legitimately, that a reading should inform rather than compel. The atlas can observe which mechanism exists; it cannot always tell whether the gap was a choice or an oversight.
• Survivorship cuts through this. Automatic systems that work quietly generate no coverage. The ones that reach the record tend to be those that failed publicly or were announced loudly.
• The counterfactual is missing almost everywhere. Only Mexico City has peer-reviewed evidence about whether its trigger achieved anything. For most entries here, nobody has measured whether the consequence helped.

The verdict

If a city is about to buy a monitoring system, the useful question is not what the sensors can detect. It is a sequence of four questions, and the atlas suggests most procurements answer only the first.

1. What number are we producing, and who sees it? Nearly every project clears this bar.
2. What happens when it crosses a line? If the answer is "we will consider it", you have bought rung one and should say so out loud rather than describe it as an early-warning system.
3. Who holds the power to make that happen, and do we hold it? Bangkok can close its own schools and cannot reach an employer. Busan can see the underpass and cannot close it. Miami lowered its threshold and lost the power to attach a duty. This is the question that decides the outcome, and it is a legal question.
4. Is the lever any good? Mexico City's trigger is the most disciplined in the atlas and pulls a lever that peer-reviewed work says does not work.

The pattern across three continents is consistent enough to state plainly: sensing capacity has raced ahead of the authority to act on it, and the gap is widening because sensing is procurable and authority is not. A rain gauge can be ordered. A barrier at an underpass needs capital works. A duty on an employer needs a law. A cut in agricultural burning across a national border needs a treaty. The sector sells the first and quietly implies the rest.

The cheapest honest improvement is not a better sensor. It is writing the consequence into the same document as the procurement, and naming who is obliged to deliver it.

One caveat sits upstream of everything here. This piece assumes the reading is sound, and often it is not: a complaint map measures who felt able to complain, not where a city is broken. Who complains takes that question apart. Downstream, our ownership essay asks who ends up controlling these systems, what survives when aid ends asks what happens when the money came from outside, and why smart cities fail collects the recurring shapes of the failure itself.