ENGINE / 01
Forecasting electoral outcomes before they happen.
Multivariate models trained on 12 years of Sri Lankan electoral, demographic, and behavioural signal — surfacing swing turnout 90 days before polls open.
Data Layer
12 years · 160+ districts
Historical Electoral Corpus
- Complete district-level results from 2010–2024 parliamentary, presidential, and provincial elections
- Demographic overlays: age, ethnicity, religion, income, urbanisation
- Voter registration deltas and turnout patterns across cycles
Model Layer
Bayesian + ensemble
Multivariate Inference
- Bayesian hierarchical models capturing district-level heterogeneity
- Gradient-boosted ensemble for swing-seat classification
- Time-series decomposition of turnout trends with seasonal correction
- Confidence intervals and uncertainty quantification on every forecast
Real-Time Inputs
- Social media sentiment velocity (trilingual NLP)
- News cycle intensity scoring
- Economic indicators: inflation, employment, commodity prices
- Polling data integration with methodological weighting
Output Layer
Operator dashboards
Campaign Intelligence
- District-level win-probability maps updated weekly
- Swing-voter segment identification and targeting recommendations
- Resource allocation optimiser: where to spend the next dollar
- Scenario modelling: what-if simulations for strategy pivots