Each poll receives a composite weight from five factors: recency (60-day half-life), sample size, population type (LV/RV/A), polling mode, and pollster transparency tier.
Win probabilities are derived from 40,000-iteration Monte Carlo simulations with stored random seeds for reproducibility. National and contest-level systematic error terms are calibrated against historical performance.
Each candidate claim is independently scored 0-100. Composite scores use a logistic mean weighted by claim recency, audience reach, and source reliability tier.
Voter survey responses feed a logistic regression model that produces marginal effects and swing scores per issue, predicting how shifting issue salience moves vote intent.