As Election Day comes to a close and results start to roll in, it can be difficult to understand what the numbers you’re seeing might be saying about who will ultimately win a key race.

CNN Sightline, a new election tool designed to surface insights from CNN’s Decision Desk, can translate those results into an understandable, real-time read on what’s happening and who is favored to win.

The tool interprets election data through modeling and assessments based on the questions that CNN’s Decision Desk analysts ask ourselves as we’re assessing the state of each major race: Where did the votes come from? How do they compare to past elections and pre-election expectations? How much statistical confidence do we have in our estimates? Were they Election Day, by-mail or early in-person votes? And what do we know about what’s left to count?

Sightline brings together the results of a range of estimates meant to answer each of those questions, shows how the CNN Decision Desk interprets results to answer those questions at any given moment and produces an overall rating on whether a race leans toward either candidate.

This year, Sightline will be live for about 25 major races for Senate and governor across the country, illuminating where things stand in the contests that will decide which party will control the Senate and which will lead critical states. Here’s a look at how it works. Sightline’s overall estimate is derived from signals in five key areas:

Outstanding vote: These models and calculations assess how much of the uncounted vote a trailing candidate needs to win, and if it’s possible based on their performance so far and where those uncounted votes are. This indicator kicks in once 25% of a state’s expected vote is reported.

Geography: This indicator reads whether the reported votes are evenly distributed throughout the state, both geographically and by partisan lean. If those distributions are not balanced, it assesses whether rebalancing would shift the final results. This indicator will start to calculate once 10% of a state’s expected vote is reported.

Comparison to past: This category starts the night looking at a range of pre-election assessments of the race, including race ratings, polling and — for later poll closing states — signals of the national environment from states that have begun to report votes. It evolves through the night to lean more heavily on a direct comparison to 2024 performance for counties that have reported a meaningful share of their expected votes. This indicator begins before polls close with a rating based on pre-election estimates. Signals from the national environment kick in as we see results from across the country. State-specific vote-driven estimators within the model will begin to contribute once 10% of a state’s expected vote is reported.

Size of lead: Will a lead hold up? This rating takes a look at the size of a leading candidate’s advantage across several models and assesses how much confidence we have that those leads will hold, factoring in how much vote has been reported. This indicator will start to calculate once 10% of a state’s expected vote is reported.

Vote method: In states where vote methods — such as votes on Election Day, by mail or early-in-person — are reported for all counties, this model estimates how well represented each type of vote is, which candidate stands to benefit when any underrepresented methods catch up, and extrapolates what the results might look like when the count is complete. This indicator will start to calculate once 10% of a state’s expected vote is reported.

Each of these components includes multiple ways of looking at the data to answer our key questions, and the relative importance of those components can change throughout the night. Sightline will move in response not only to updated vote count, but as our ability to assess the vote in different ways changes.

For example, in the Vote Method indicator, early in the night, when vote reporting is most uneven, we put more weight on what we know about how well represented each method is and how voters who used that method split between Democrats and Republicans in the past. As the vote count builds and our models based on what votes by method are telling us become more precise, that part of the scale becomes a bigger driver of the forecast.

Sightline’s forecasts are not the same as a formal CNN projection of a winner in a race. The indicators underlying the overall assessment might all point in one direction, or sometimes they may send opposing signals. Either way, they provide a look at what CNN’s experts at the Decision Desk see in the data, but they can and will change as what we know about the race grows.