CivicLoom, a Singapore-based urban data company, said on Saturday that its forecasting software overstated projected 2035 populations by a combined 1.6 million people across 14 cities in eight countries. The company will refund S$2.4 million (about €1.6 million) in fees and has begun notifying planning teams whose reports used the figures.
The error surfaced after CivicLoom published a decade of archived software code for an anniversary exhibition on 17 October. Niyati Sarma-Gopal, a researcher at the fictional Lion City Institute of Digital Affairs, reproduced one of the old forecasts and found that a small adjustment intended to apply once had been added to every year in the projection.
The code behind the numbers
The adjustment, labelled a “recovery allowance”, added 0.8 percentage points to annual population growth. CivicLoom said it was meant to soften projections after a temporary dip in migration, but a coding choice applied it repeatedly across forecast years. In affected cases, the resulting 2035 estimates were 4% to 11% higher than corrected calculations.
Sarma-Gopal said she checked the result against the company’s archived examples and reproduced it on a separate computer. A public change log showed that engineers had flagged the repeated allowance in 2019; the entry was closed with a note describing it as acceptable scenario behaviour. “The surprise was not that the model had assumptions,” she said. “It was that this one kept getting another turn.”
We shipped a map that looked more certain than the cities behind it.
Emanuel Abramović, former CivicLoom product engineer
Clients revise their plans
Abramović, who left the company in 2022, said the allowance had been added during an early product test and was intended to represent a single rebound after a weak year. He said he raised concerns that the setting compounded, but did not know whether its effect had been checked across all client forecasts.
CivicLoom chief operating officer Tatiane Lovato said the company found the issue while preparing the anniversary archive, then asked an outside modelling team to review its forecasts. “We should have caught this before those projections went into client work,” she said. The company said it had not found evidence that staff altered figures for any particular city or customer.
Six planning teams used affected forecasts in transport, housing or utility proposals, CivicLoom said. None has reported a project being built solely on the inflated numbers, but three are revising procurement estimates. The company has offered corrected files and says it will publish a city-by-city impact table by 15 November.
The episode has renewed debate over population forecasts used in fast-growing cities, where a few percentage points can affect choices about rail capacity, water systems and land for housing. Singapore’s own population statistics were not affected; CivicLoom’s models are commercial planning tools, not official counts.
At the anniversary exhibition, the archived code now sits beside an interactive map showing the corrected curves. Lovato said the company would keep the archive online and add a plain-language explanation of its assumptions. “The anniversary was meant to celebrate ten years of better city planning,” she said. “It has become a useful lesson in showing the workings, too.”
