Market vs. Desk

Archive: the 2026 midterms tracker no longer updates. Scores include every call logged; race charts show election night.

How the numbers work

The rules behind the scoreboard, and why some counts on the page differ. Back to the scoreboard

When does a market “call” a race?

A market calls a race once its odds hold at 95% or more for one candidate for five straight minutes. We run that rule three ways: on Kalshi alone, on Polymarket alone, and on the average of the two. The average is the number the main scoreboard uses.

The call is stamped at the start of that 95% run, since that is when the money moved. Many safe seats sat above 95% all day, so their market call is stamped hours before polls closed.

Why 95% and not 99%?

Because 95% held for five minutes has worked as near-certainty. Of the roughly 430 races both a market and a desk called, the Kalshi/Polymarket average picked the wrong winner in 4, about 1%.

A higher bar mostly measures the order book, not belief. Prices rarely settle at 99¢ or $1 even when an outcome is certain, because buying the last cent isn't worth the fees and the money tied up until the race settles. The site reads the midpoint between the best bid and the best ask, which often sits near 98.5% on a decided race. Kalshi's own 99% figure is different again: it counts the first single trade at 99¢ or more.

Replaying election night's odds under each threshold shows the trade-off between speed and accuracy (the Kalshi/Polymarket average, against the earliest desk call):

ThresholdMarkets aheadDesks aheadTiesWrong winner
95% (used here)207 (1h 46m)11 (9m)2114
97%186 (1h 21m)38 (9m)2063
99%53 (52m)301 (51m)760

Counts of races, with the median lead in parentheses. Replayed from stored odds on the morning of Nov. 4, so they differ slightly from the live scoreboard. “Wrong winner” counts races a desk called for the other candidate or where the market's pick later fell below 50%. At 99%, three desk-called races never get a market call at all.

The cost of 95% is a handful of early calls that later reversed. The site counts those as misses and shows them, so the trade-off stays in view.

Why do leads count from poll close?

A market can know a safe seat's winner days ahead, but a desk will not call any race until its polls close. If we counted from the moment the market reached 95%, every safe seat would show a lead of many hours and swamp the numbers. So a market call made before polls closed counts from poll close. Those calls show as “before close” or “(poll close)” on the site.

Poll close is set per district where a state spans two time zones. Most of Florida closes at 7 PM ET while the panhandle (FL-01, FL-02) closes at 8. The Eastern-time districts of Kentucky and Indiana close at 6 PM ET. Most of Michigan, Texas and Kansas close an hour before their westernmost counties, and ID-02 closes an hour before ID-01. Statewide Senate races use the state's last closing. Chamber control counts from 7 PM ET.

Why are so many races “tied”?

Calls less than a minute apart count as a tie. We read market odds once a minute, so a market's call time is only accurate to the minute, and a few seconds' difference means nothing. Most ties are safe seats: the markets were already at 95% when polls closed, and DDHQ called the race at the same moment.

How is the median lead measured?

The median lead counts only races where one side, a market or a desk, was more than a minute ahead. Ties are left out. If they were counted, the hundreds of safe seats called at poll close would pull the median to zero and hide how far ahead the markets were in the races where timing actually differed.

The paragraph at the top of the page gives the full picture for each market: how many races it beat the desks on and by how much, how many the desks won and by how much, how many were ties, and how many it called wrong.

What counts as a miss, and why don’t the miss counts add up?

A market misses a race in either of two cases:

  • a desk calls the other candidate, or
  • after calling the race, the market's pick falls below 50%.

The second case is strict. A market that wobbles below 50% overnight still counts as a miss, even if the desk later calls the same candidate the market first picked.

Three places on the page count misses, and they count different things. The scoreboard at the top counts races where the Kalshi/Polymarket average missed, but only races a desk has called. The scorecard lower down counts the same average's misses in every race, including reversals in races no desk has called yet. The Kalshi vs. Polymarket table counts each market on its own: when both markets miss the same race, it counts once in the scorecard and once in each column, so the two columns can add up to more than the scorecard's total.

How are Kalshi and Polymarket compared?

Each market is scored on its own odds with the same 95%, five-minute rule. “Called first” only counts races both markets list, since Polymarket lists almost every House district and Kalshi lists the most-watched ones. That is also why Polymarket has far more calls in total.

Why don’t Kalshi’s published numbers match this site’s?

Kalshi published its own scorecard after the election, and it measures differently. It times a Senate call from Kalshi's first trade at 99¢ or higher and compares that with CNN's projection. This site times a market call from when the Kalshi/Polymarket average first held 95% for five minutes, counts it from poll close at the earliest, and compares it with the earliest call from DDHQ or AP.

Kalshi's accuracy figures also cover a different set of races, including governor and local races this site does not track, and they take prices at fixed moments such as when polls opened. Its numbers are reproduced as Kalshi published them, with a link to the source.

Where do desk calls and their times come from?

We track two decision desks: Decision Desk HQ and the Associated Press. When both call a race, the earlier call is the one the markets are scored against. If they call it at the same moment, DDHQ counts.

Calls arrive from several places and are refined as better times come in. DDHQ's posts on Bluesky carry its call time to the minute. AP's Bluesky posts carry no call time, so we use the post time until a better one arrives. A race-call ledger then supplies each desk's published call time to the hundredth of a second. A few calls were entered by hand.

What counts as a flipped seat?

A flip is a race called for the party that does not hold the seat now. “Desk called” means a desk made that call; “markets only” means the markets have called it but no desk has yet.

The holder is the party of the seat's current member. House seats go by district number, so a district redrawn since the last election is counted against whoever represents that number today. Vacant seats are left out.

How are the Senate and House totals and projections worked out?

The large numbers are seats a desk has called. In the Senate they include the 65 seats not on the ballot this year (34 held by Democrats and the independents who caucus with them, 31 by Republicans). The light part of each bar is seats only the markets have called.

The projection keeps every desk call and gives every other seat to the candidate the markets currently favor. It notes how many uncalled races are still within 35–65%, where the favorite could easily lose.

A few races are not tracked because they are not a straight Democrat-vs-Republican contest in the markets: NE-Sen, AK-AL and CA-40. They are left out of the totals.

How is the paragraph at the top written?

It is generated from the same numbers as the scoreboard and the Kalshi vs. Polymarket table and updates as calls come in. “Races left to call by markets” counts tracked races where no market call is standing; “left to call by desks” counts tracked races no desk has called.