Election methodology · election-accountability-v1

One board. One minimum. Every receipt.

These rules govern people, forecasters, models, and newsrooms identically inside the same election cohort.

What counts as a prediction.

Lean, Likely, and Solid all select the named side. Toss-Up and Pass make no directional prediction. Correct and incorrect calls settle against certified results; live projections remain provisional.

Who appears.

An identity must make directional calls on at least 80% of the exact displayed slate before cutoff. Below that standard, it does not appear on the board. Historical success and roster admission never create a current-cycle row.

How rank works.

Eligible records sort by the Wilson 95% lower bound, then larger resolved sample, then fewer misses on the predeclared competitive set. Identity type never modifies rank.

How evidence enters.

Forecast artifacts remain immutable, source labels are normalized under versioned rules, and human review must admit extracted entries before they affect public aggregates. Corrections append; they never erase.

What settles the record.

Associated Press or the declared alternative may power clearly labeled provisional calls. State and local certification records are authoritative for final scoring.

Open the election board →