Indonesia

What the record shows

Indonesian cities have a habit that is rarer in this atlas than it ought to be: the operator publishes the figure that spoils its own announcement. Ahead of the 2025/26 rainy season, Bandung set out what its flood early-warning system actually consists of — thirteen cameras watching rivers and channels, and water-level sensors at two locations. Two sensors is a defensible starting point and a very thin instrument for a basin city, and the useful part is that the city answered in numbers rather than adjectives. Read the Indonesian entries together and that habit keeps reappearing, almost always attached to the same gap: the distance between what a system is capable of and what it is actually doing.

The clearest case is a mayor contradicting his own procurement. Bandung's area traffic control system runs signal controllers that can retime themselves from live counts. In July 2025 Mayor Muhammad Farhan said the adaptive capability is installed and unused, for want of the counts it needs. What the system does produce is a public violation dashboard: 370 recorded offences at eight junctions in the first six days of August 2026, involving 248 vehicles, 208 of them two-wheelers. Cameras that observe and tally, controllers that could act and do not — the equipment is adaptive, the operation is not. That distinction is worth carrying into every adaptive-signal claim anywhere: capability is bought, operation is staffed.

The same gap turns up in gentler forms. The MRT that opened in March 2019 is driven automatically by its signalling and still carries a driver in the cab — GoA2, not driverless, and the entry says so rather than letting the technology name imply the outcome. Bandung's BOSEH bike share is the inverted version: the automated docking that makes a bike share cheap to run has degraded to the point where the transport department trucks about 200 bikes out to 23 active shelters each morning and collects them again at night, from a network that started at roughly thirty sites. Automation was bought; a van, a driver and a booth attendant now do its job. Spare parts, not launch fleets, decide whether a docked scheme survives its first decade.

Where measurement is genuinely good, the country's record is unusually candid about what measurement does not buy. Thirty-two residents sued the president, three ministers and three governors over Jakarta's air; the court found all seven had acted unlawfully in September 2021, and the province's network grew from five reference stations to 111 monitoring points feeding a public index. But no index threshold carries an automatic legal consequence. A court can compel a city to measure and cannot compel it to act on what it measures — the sharpest illustration in the atlas of why a reading and a decision are different purchases. Bantargebang makes the point physically: every truck has been weighed in and out on networked bridges since 2015, with GPS flagging vehicles ten kilometres before the gate, and the daily tonnage published. A waste slope collapsed on 8 March 2026 and killed seven people. Metering the gate is not the same as reducing the load.

Enforcement fails at the least technical link. Jakarta's camera enforcement has run since November 2018 and reads plates well; the notices then go out by post to the address on the vehicle registration. In 2023 police could post about 800 of roughly 12,000 daily violations on their postage budget, and the traffic director put the confirmation rate at 0.01 per cent. That number matters beyond Indonesia, because it is the half of automated enforcement no procurement covers: a camera network is only as strong as the address database and the postage budget behind it. It also decides the fate of what the cameras are pointed at — odd-even plate rationing, in place since 2016 on 26 corridors, which a regression-discontinuity study found does not significantly cut the city's air pollutants, while the priced alternative promised since the 2000s has survived tenders and courts without ever opening a gantry.

The administrative layer tells a quieter story about inheritance. Bandung spent an era launching a separate app per service and now runs Sadayana as a single front door, with its communications department stating plainly that the city is no longer competing to build more apps but integrating the ones it has. Its transport unit made the same choice in miniature, putting city buses, tour buses and school buses on one tracking map rather than one app each, and — unable to mandate a fare medium — simply made cash cost a quarter more than the QR alternative. Surabaya reached the same destination two decades earlier and from the opposite direction, running budget and procurement through one integrated system since 2004 and permits through a single window since 2013. What none of these publish is use: Bandung's open data portal is genuinely maintained, carries 2,083 datasets, and its own API reports a median of six downloads per dataset, with 187 never downloaded at all. Counting datasets flatters a portal.

Two entries show contract design doing work that technology is usually credited with. TransJakarta absorbed the informal angkot minibuses by paying operators per kilometre driven rather than per passenger, which removes the incentive to race — and shifts quality control onto the buyer, who logged 827 crashes in the first nine months of 2022. KJP Plus matched its school-allowance roll against tax and asset registries, stripped 105,225 pupils of their cards, and was politically reversed within months. Against that, Nusantara is the outlier the others explain: a capital designed around driverless shuttles and an eighty-per-cent-renewables target for 2045, currently at a fifty-megawatt solar plant covering roughly a tenth of its power, with the 2026 state budget cut 85 per cent from 2024. Indonesia's strongest entries are the ones that count what exists. Its weakest is the one still counting what is planned.

Sources beyond the atlas

Every factual claim above is carried by one of the linked atlas entries; no external source was needed for this section.