Ranked-Choice, Top-Two, Runoffs and Fusion: How State Election Rules Change the Way a Race Is Modelled

A U.S. election is not one kind of contest held 51 times. Each state decides how candidates reach the November ballot, how many winners a district elects, what share of the vote wins, and which year a chamber is even up. A forecast or a results page that treats every race as a plurality contest between one Democrat and one Republican will be wrong in predictable places.

When we built 99 Chambers, a forecast of every state legislative seat plus every Senate, House and governor race in 2026, we put each state's rules into a registry: one file per state and DC, read by the loaders, the model and the publisher. This article walks through the rules that most change how a race has to be modelled and reported, with the examples we encoded.

Ranked-Choice Voting: Model the Final Round

Under ranked-choice voting (RCV), voters rank candidates. If nobody has a majority of first choices, last-place candidates are eliminated and their votes move to each voter's next choice until a winner emerges. Maine's election law defines it as tabulation "in sequential rounds in which last-place candidates are defeated" (21-A §1).

Where RCV applies differs a lot. Alaska uses it in November for every office, legislature included. Maine uses it in general elections only for U.S. Senate and U.S. House; the statute lists primaries for governor and the legislature, but not their general elections, so those are decided by plurality. The District of Columbia uses it from 2026 in races with three or more candidates, with up to five rankings.

For modelling, the quantity that matters is the final two-candidate margin, not first-choice shares. A first-round lead below 50% is not a win, and reporting it as one misleads readers on election night. We model the final round and say so on the page.

Top-Two and Top-Four Primaries: Same-Party Finals

In a top-two primary, every candidate runs on one ballot and the two with the most votes advance. California's Secretary of State puts it plainly: the top two "move on to the general election regardless of party preference" (CA SOS). NCSL lists California and Washington as top-two states, Alaska as top-four, and Nebraska's nonpartisan legislature as a close variant with no party labels on the ballot (NCSL).

The consequence is that November can feature two Democrats or two Republicans. When that happens, party control of the seat is already settled, and a model that simulates a two-party margin there is producing noise. Our registry marks these states, and a race with two same-party finalists is fixed for that party and shown as a lockout. In Nebraska, where ballots carry no party, we use each candidate's registered party to say which side a seat counts toward.

Runoffs: Report the Chance of a Second Round

NCSL notes that Georgia, Louisiana and Mississippi require runoffs in general elections, and that in Georgia a runoff "must be held if no candidate receives more than 50% of the vote" (NCSL). In Georgia that rule covers federal, statewide and legislative races, and the 2026 runoff date is December 1.

A runoff state needs a second output alongside the win probability: how likely the race is to go to another round at all. Our simulation checks whether either candidate clears the threshold, given the third-party share. If not, the margin gets extra runoff uncertainty, and the page shows the chance of a runoff as a separate gauge. Rules also move between cycles. For 2026, Louisiana put its U.S. House races on a November 3 all-party ballot with a December 12 runoff, while its Senate race went through closed party primaries. Because that lives in the registry, with earlier cycles recorded separately, our 2022 backtest still replays 2022 under 2022's rules.

Fusion Voting: Add the Lines, Keep Them Visible

Fusion lets more than one party nominate the same candidate. Ballotpedia describes two forms: full fusion, where a candidate appears once per party line and the lines are added together (Connecticut and New York among them), and an aggregated form in Oregon and Vermont, where one name is listed with all its nominations (Ballotpedia).

Our registry marks fusion in New York, Connecticut, Oregon and Vermont. In New York and Connecticut we sum a candidate's lines and keep each line for display, because a reader comparing our numbers to official returns will see a Working Families or Conservative line separately. In Oregon and Vermont the candidate's votes count once under their main or first-listed party. Vermont adds one more wrinkle: its Progressive Party caucuses with the Democrats, so the registry lists it as allied and its seats count toward that side's majority.

Multi-Member Districts: Seats, Not Just Winners

In several states a district elects more than one legislator. Our registry flags multi-member chambers in Arizona, Idaho, Maryland, New Hampshire, North Dakota, South Dakota, Vermont, Washington and New Jersey. They do not all work the same way. Washington elects two House members per district but as separate positions, so each position is its own race. New Hampshire elects a 400-member House from named districts where voters pick as many candidates as there are seats, plus floterial districts that overlay others. DC's at-large Council seats cap how many one party can win.

The right output for a multi-member district is expected seats per party, not a single winner. We model each seat's margin as varying around the district's, and store expected seats per district.

Off-Year and Staggered Chambers: Who Is Not on the Ballot

Not every chamber is up in a midterm. Virginia, New Jersey, Louisiana and Mississippi elect their legislatures in odd years, so none of their legislative seats are on the 2026 ballot. Many senates are staggered: California's Senate puts the even-numbered districts up in 2026, and Alaska's Senate elects half its seats. Michigan, by contrast, elects its whole Senate in midterm years.

Chamber control depends on the seats not up as much as the ones that are. We count holdover seats at the current members' party split and add them to every simulation. Control also depends on tie rules, which the registry records per chamber: in some states the lieutenant governor breaks ties, so a tie can hinge on a separate statewide race, while Minnesota's House has no tie-breaker at all.

What This Means for Your Reporting

• Write each rule down as data, per state and per office, rather than as exceptions scattered through code or spreadsheets.
• Report the number that matches the rule: final-round margins for RCV, runoff chances for majority states, expected seats for multi-member districts.
• Flag same-party finals as settled for party control instead of simulating them.
• Keep past cycles' rules alongside the current ones, so comparisons and backtests stay honest.
• Explain each rule on the page in one or two neutral sentences, so readers understand why a race looks different.