Forecast methodology
A forecast is a conditional estimate, not an election result.
Prepare verified inputs
Confirm the election timetable, official candidates, source references, geographic identifiers, reviewed survey responses and sample design. Historical results remain separate from predictions.
Execute a reproducible model
The proposed initial approach combines weighted survey estimates with documented historical adjustment. A frozen input snapshot and model version identify every run. This statistical engine is not yet implemented or calibrated.
Validate before approval
Check candidate eligibility, duplicate responses, source freshness, sample coverage, vote-share totals, interval bounds and LGA-to-state aggregation. Failed or insufficient-evidence runs cannot be published.
Publish approved aggregates
Publication is currently disabled at the database level. A future approval process must release aggregate estimates and uncertainty together, preserve the run record, and write a privacy-safe audit event.
Evidence before estimates. Only reviewed, source-backed data can support a published forecast.
Our methodology