Every simulator is built so an election result is reproducible from real inputs using the country's actual electoral law, not a black box. Each simulator starts from a verified baseline: the most recent real election entered at the finest geographic level we can source (commune, constituency, county, district, province or region), taken from official electoral authorities and cross-checked against published national totals, and carried on real boundary data.
Redistributing support. National changes are spread across units with a swing model, not by flattening every unit to the average. Uniform swing moves each unit by the same points as the national shift (s' = s + (S' − S)); proportional swing scales each unit by the national swing ratio and renormalises; bloc-preserving swing swings whole ideological families so the balance within a bloc in each place is preserved and only the balance between blocs moves.
Spatial vote projection. For custom candidates with no historical geography, each real baseline voter flows to the ideologically nearest candidate by a softmax over Euclidean distance, P(v→c) = exp(−λ·d(v,c)) / Σ exp(−λ·d(v,k)), then national totals are calibrated to the chosen shares by an iterated per-candidate multiplier.
Seats. Each allocation family is implemented exactly as legislated. Highest-averages: D'Hondt uses quotient V/(s+1), Sainte-Laguë uses V/(2s+1), modified Sainte-Laguë uses a first divisor of 1.4. Largest remainder uses the Hare quota V/S or the Droop quota. Legal thresholds (5% in Germany and Poland, 4% in Sweden and Norway, and others) remove parties before allocation, with statutory exemptions honoured. Two-tier systems add national levelling seats so each eligible party's total seat share matches its vote share; Estonia's modified-D'Hondt compensation exponent is applied exactly. Apparentement and Bader-Ofer surplus-sharing agreements are honoured. Mixed systems combine a single-member tier and a list tier by parallel voting (added independently, e.g. Japan) or compensatory MMP (list top-up toward proportional entitlement with overhang, e.g. Germany and New Zealand).
Runoffs. Eliminated candidates' voters are redistributed between the two finalists by ideological proximity, with abstention rising for voters far from both, then the national two-way split is calibrated to the published runoff poll by a uniform odds-ratio factor.
Election night. Reporting units are revealed in randomly-sized batches on a Gaussian timing profile; seat projections and the dashboard are computed only from units that have actually reported, never by extrapolating the final number backward. Validation: before release, each baseline must reproduce the official certified result exactly under our own allocation code. Projected scenarios are built from public polling aggregates and framed as scenarios, not forecasts.