Cases

In Korea’s 9th nationwide local elections (2026), Alea Capta forecast the field and vote shares with ontology, synthetic population and a real-time forecasting system, coming within 0.4 percentage points of the actual count in Seoul and calling 14 of 16 metropolitan races (87.5%), verified after the fact with National Election Commission count data. It shows what happens when a domain is put on our engines.

Case · Politics and elections

We read the election before the count.

The challenge

The field moves every day during an election. Phone polls are expensive and slow, and once asked they cannot be re-run under different conditions. What the client needed was to know, every day, where the race stood, how the count would turn out, and what would change it.

The approach · how the engines ran

01Ontology
We laid out candidates, parties, regions, issues, events and their causal links as a political-domain knowledge graph.
02Synthetic population
From Statistics Korea data we reproduced an electorate of 1 million people with about 170 attributes each, down to the district and neighborhood level.
03Calibration
We compared past events with real shifts in voter sentiment to score each event’s influence, and re-tuned every week by matching forecasts against outcomes.
04Real-time forecast
Each time a new event came in, Monte Carlo produced a new distribution of vote shares; every forecast was recorded in a ledger and compared with actual results.
05Delivery
We delivered dashboards and conversational AI that give evidence-backed answers, and videos that explain policies.

Results

Flagship · AIPP

In Korea’s 9th nationwide local elections in 2026, our AI election forecasting system AIPP proved its accuracy against real results. Instead of polling, it built a synthetic population of 1 million people reflecting real demographics, asked them directly, and recalculated the field and vote shares every day with an ontology and a real-time forecasting system.

0.4%p
Seoul forecast vs. actual count
Most precise
±0.5%p
Average error in key battlegrounds
Battleground avg. error
87.5%
14 of 16 metropolitan races called
Metro mayor hit rate

Verified after the fact with National Election Commission count data
Reviewed by policy institutes of major Korean parties
Listed as an exemplary case in the election white paper

Also built on the same engines

Conversational AI

Policy chatbot

We built and deployed a conversational AI for election campaigns that explains every pledge of a candidate and answers citizens’ questions in real time.

2,500conversations in 5 days

Campaign Ops

AI campaign operations app

For local elections nationwide, we built and delivered an app that generates and manages schedules, optimizes routes, and drafts pledges and speeches for each campaign stop.

20+election campaigns

Policy Video

Policy promotion videos

We modeled real places from photos and coordinates and explained pledges in 30–90-second vertical videos with figures, narration and subtitles.