Sunday, 8 November 2026 The Plausible Post Independent news from around the world, around the clock.

Software Glitch Exposes Inflated Accuracy Claims in Philippine Drought App

A software accident has revealed that a Philippine start-up’s drought-prediction app counted duplicated rain readings as separate proof of its accuracy. The company won ₱48 million (about €740,000) in public contracts on the strength of forecasts it said were right 96% of the time.

By Scot Bernhard ·

A municipal agriculture planner checks a rainfall-risk map in Iloilo City after a software glitch exposed how the app calculated its accuracy score.
A municipal agriculture planner checks a rainfall-risk map in Iloilo City after a software glitch exposed how the app calculated its accuracy score.

Amihan Gridworks’ HabiCast app was built to help local agriculture offices prepare for dry spells, including the El Niño conditions forecast to intensify before December. But a faulty software update on 24 September briefly put a testing layer on the public map — and exposed a scoring method that made the forecasts look far more reliable than they were.

The company said the mix-up lasted 38 minutes and affected no water controls or farm equipment. A technical review commissioned by three provincial agriculture offices, shared with The Plausible Post on Thursday, found that the app’s advertised 96% accuracy rested on repeated readings from the same rain gauges being counted as separate checks.

The map that showed too much

The map normally displays a colour-coded estimate of rainfall risk across towns in the central Philippines. During the accidental rollout, users also saw clusters of identical gauge markers over parts of Iloilo and Negros Occidental, including readings stamped with the same minute and measurement.

“We assumed each dot represented a separate observation. Then we noticed the same gauge code repeated across the map,” said Martim Soares, a data analyst with one of the provincial offices. His team reported the display to Amihan; the company disabled the layer before lunchtime.

The HabiCast map briefly displayed repeated gauge markers during a software update in September.
The HabiCast map briefly displayed repeated gauge markers during a software update in September.

The review says the app copied a single station’s reading into as many as 18 nearby map cells, then counted those copies as independent comparisons when calculating its score. That could make one correct observation appear to be a string of successful predictions. The review found no evidence the company changed rainfall records themselves, but said its headline accuracy figure could not be independently reproduced.

Contracts face a fresh test

Amihan’s data director, Gael Sanches, said the duplicated points came from an effort to make patchy station coverage easier to display. “We got the map logic wrong, and then we used a score that rewarded the same reading more than once,” he said. The company has withdrawn the 96% claim and offered the offices a new evaluation using distinct gauge records.

HabiCast was introduced last year to give town agriculture teams a simpler way to turn seasonal outlooks into local planning advice. The offices paid a combined ₱48 million (about €740,000) for access, training and forecast support; they are now checking payment milestones before deciding whether to extend the contracts.

For farmers, the app was an extra source of guidance rather than an automatic instruction to change planting plans. Viviane Delgado, who coordinates a growers’ group outside Iloilo City, said members checked its alerts against local rainfall and advice from agriculture staff. “It was handy when the screen worked. We still kept our notebooks and watched the sky,” she said.

The wider seasonal picture remains important: a forecast issued by the World Climate Council expects El Niño to strengthen before reaching its peak in December. Amihan says it will publish its revised scoring method by 20 October and let the provincial offices test it against past seasons before any further accuracy claim appears.

Officials said the app will remain available while the review continues, but its forecast score has been removed from the service. The accidental map display has turned a routine software patch into a public test of whether a useful local tool can earn trust back one properly counted rain gauge at a time.

More news