# Polling-only governor forecast charter

This charter covers the Wisconsin, Nevada, Georgia, and Michigan 2026 governor research previews. It does not replace the Oregon rematch charter.

## What the probability means

Each displayed percentage is a conditional modeled chance of winning among the two modeled major-party candidates, not projected vote share and not an unconditional probability over every ballot option. The aggregate remainder is assigned a vote-share distribution but no win probability; its public value is `null`, never a numeric zero. The center of the distribution is the quality-, recency-, sponsor-, and repeated-pollster-adjusted average of reviewed public polls. Election-day uncertainty comes from the nearest fixed-horizon record in the repository's out-of-time gubernatorial polling-error calibration.

The prior governor result and the certified 2024 presidential result are descriptive comparisons only. Their midpoint is displayed on the evidence chart with `forecast_effect: 0`; it does not pull the polling estimate in either direction. Reviewed political context also has exactly zero forecast effect.

## Scope

- Exactly two major-party candidates are modeled.
- Minor, independent, write-in, and undecided responses remain in the aggregate remainder unless a race-specific note says otherwise.
- The major-candidate margin is the difference between their simulated all-ballot vote shares. A separate normalized two-candidate margin is retained only for like-for-like evidence comparisons.
- The pollster influence cap applies only to repeat polling by one organization. Two singleton pollsters retain evidence-proportional recency and quality weights; the cap never turns an old weak poll into an equal-weight poll by construction.
- The release status remains `research_preview`.
- The same source, input-hash, immutable-archive, and Analyst's Take boundaries used by the Oregon page apply to every race.

Georgia requires a majority to win. Its research preview reports eventual two-major-candidate win chances under the explicit simplifying assumption that the simulated first-round major-party leader also wins any runoff. It separately reports a descriptive runoff probability and does not simulate second-round campaign movement.

Nevada includes a named nonpartisan candidate and the non-winning “None of these candidates” option. This first preview aggregates those options rather than assigning them individual win probabilities. That limitation is disclosed in the artifact and page warnings.

## Update rule

A new run may enter publication only when its dated inputs pass validation, its embedded hash matches the exact public JSON, and the immutable archive byte-matches the public artifact. A correction produces a new run; it does not rewrite an archive.