Kalshi listed its 2028 House contracts in May 2026, more than two years before anyone votes, and the traders on that board are already pricing against the steadiest pattern in modern American politics. Since 1946 the House of Representatives has changed hands in a presidential election year exactly twice, in 1948 and in 1952. Eighteen straight presidential elections have passed without one, a run of 72 years. Traders on this board are paying about 34 cents on the dollar for Republican control after 2028, and most of the path to that outcome runs straight through the thing that has not happened since Harry Truman was president.
We asked eight AI models the same question with the price hidden from all of them, handed each one the full postwar control ledger, the live polling aggregates and the enacted-map record, and let them read each other's reasoning and revise. Where they landed, and the one column on their scorecard where seven of eight quietly broke with the history book, is below. First, the two things the board was priced on: the map fight that is still in court, and the record every seat was told to read before it wrote a number.
This is a neutral, data-driven look at market prices, the public record and historical patterns; it does not advocate for any party, candidate or outcome.
Stay ahead of the markets.
Daily insights and expert picks on Kalshi, Polymarket, and what's moving markets.
Free forever. Unsubscribe anytime.
The Map Fight That Is Still In Court
Most of the districts these races will be run in were redrawn during a mid-decade war that reached the Supreme Court twice inside five months. One side of that war is settled for 2028. The other is still in court, and which one is which matters more for this board than anything in the polls.
Texas moved first. The legislature passed HB 4, a mid-decade congressional map, on August 23, 2025, and Governor Greg Abbott signed it on August 29, drawn to shift the state's delegation from 25 Republicans and 13 Democrats toward 30 and 8. On November 18, 2025 a three-judge federal panel in the Western District of Texas blocked it, finding the redistricting "constituted a racial gerrymander". The Supreme Court stayed that injunction 6-3 on December 4, 2025, faulting the district court for having "failed to honor the presumption of legislative good faith" and "improperly inserted itself into an active primary campaign," and summarily reversed it on April 27, 2026. The map governs 2026. The lawsuit itself, LULAC v. Abbott, is still pending on the merits, with plaintiffs alleging a Section 2 Voting Rights Act violation no court has ruled on. The Texas lines for 2028 are the single largest open question on this board.
California answered at the ballot box, and it was not close. On November 4, 2025 voters approved Proposition 50 by 64.42% to 35.58%, a 28.84-point margin, under a title that named its own purpose: "Authorizes Temporary Changes to Congressional District Maps in Response to Texas' Partisan Redistricting." That measure hands map-drawing to the legislature for the 2026, 2028 and 2030 elections, after which the independent commission resumes in 2031. Federal courts declined to block it and the Supreme Court refused to pause it before the midterms.
Ten states enacted new congressional maps ahead of 2026, eight redrawn to Republican advantage and two to Democratic advantage; NBC News' tally puts Republicans in position to gain up to 16 seats from the redraws against up to six for Democrats. Measured against 2028 instead of 2026, that record cuts in an odd direction: California's advantage is locked in by its own ballot language, and Texas's is the one under active legal threat.
What The Record Says About Presidential Years
The panel's data card carried no prices and no punditry. Its spine was the full party-division ledger for every House election from 1946 to 2024, parsed at run time rather than recalled, and the base rates that fall out of it are unusually clean.
- Control Turnover Is A Midterm Event. The House changed hands in seven of the last 20 midterms: 1946, 1954, 1994, 2006, 2010, 2018 and 2022. It changed hands in two of the last 20 presidential years, both more than seven decades ago.
- The Carry-Over Rate Governs This Market. A party that flipped the House in a midterm held it at the next presidential election six times out of seven. The lone failure was 1946 into 1948, when the new Republican majority was swept out on Truman's upset and Democrats gained 75 seats.
- The Exceptions Needed Swings Nobody Produces Now. 1948 moved 75 seats in a night; 1952 moved 22, on a presidential margin of 10.85 points and 442 electoral votes. Net change in Democratic House seats at each presidential election since 1956 has been mostly single digits, with the two largest at plus 37 in 1964 and minus 34 in 1980. Reagan's 1980 landslide moved 34 seats and still did not flip the chamber.
- Winning The White House Does Not Deliver The House. In eight of the 20 presidential years since 1948 the two went to different parties, most recently in 2012.
- The Midterm-Penalty Rate Does Not Apply Here. The president's party lost House seats in 18 of 20 postwar midterms, but 2028 is a presidential year. Any model that imports that statistic into this market has made a category error, and the card said so explicitly.
Everything above turns on the carry-over rate. One public forecast, the fiftyplusone model published August 3, 2026, gives Democrats an 85% chance of winning the 2026 House with a median of 230 seats in a range of 211 to 253. Multiply that 85% by the six-of-seven carry-over rate, 85.7%, and you get 72.8%. Add the branch where Republicans hold 2026 and lose the chamber in 2028, 15% times the ledger's 2-in-20 presidential-year flip rate, and the naive total is about 74% for Democratic control after 2028. Hold that number. Neither the market nor the panel printed it, and the reason both came in under it is the whole argument of this page.
The Two Steps Every Seat Priced
Asked as a single question, this market invites a hunch. Broken into two, it becomes a pair of numbers anyone can check: who wins the November 3, 2026 midterm, and whether that winner still holds the gavel after November 7, 2028.
Step one, the 2026 midterm. All eight seats put Democrats between 82% and 85% to win the House on November 3, 2026, just under the public model's 85%. The live inputs point the same way: a six-aggregator generic-ballot average of Democrats 47.8% to Republicans 41.5% as of August 20, 2026, and presidential approval averaging 37.6% approve to 59.3% disapprove across ten aggregators as of August 17. The card also told every model to test the two midterms that broke the pattern, 1998 and 2002, against present conditions. One needed an impeachment backlash, the other a rally-around-the-flag with approval above 60; neither appears in the fetched numbers.
Step two, the hold. This is where the panel earns its keep, and where it parts company with the raw record. The historical carry-over rate is about 86%. Seven of the eight priced the hold below that in round one, and six of those seven named the same mechanism: majority size. A 253-seat majority is close to unflippable in a presidential year; a bare 218-seat majority is one retirement and two bad districts from gone, and the public model's range runs from 211, not a majority at all, up to 253. Claude Sonnet pointed instead at the map tilt, since the redraws net Republicans more seats and Texas's HB 4 governs while its case is pending. Kimi alone sat at the base rate, arguing the map litigation gives Democrats a live upside that offsets a thin majority's fragility.
Two components, then: a step-one estimate around 82% and a hold most seats put right around 80%, which multiplies out to roughly 66. The rest comes from the branch the raw ledger nearly ignores, Republicans winning 2026 and losing the chamber back in 2028: the seats priced a thin Republican majority as no safer than a thin Democratic one, a hold near 80%, so about a fifth of that 18% comes back to the Democratic side. Add the two and you land four to five points under the 74% the history book alone produces. What the market printed for the same question, and how close the eight numbers sit to it, is the board.
The Board: Market Vs. The Eight Models
| Outcome | Last trade | Book (bid / ask) | AI blend | Gap |
|---|---|---|---|---|
| Democratic House Control After 2028 | 65¢ | 65 / 70 | 69.4% | +4.4 |
| Republican House Control After 2028 | 34¢ | 6 / 34 | 30.6% | -3.4 |
The last trades on Kalshi print Democratic House control after 2028 at 65 cents and Republican control at 34 cents as of August 27, 2026, 4:40 AM ET, and the two sum to 99 because the contracts trade in separate order books. The panel lands at 69.4% for Democratic control after a revision round, the median of the eight final numbers, with every model between 64% and 76%.
Since this page's August 23 read, when the last trades sat at 72 and 28, the board has drifted about seven cents toward the Republican side: the Republican contract traded up from 28 to 34 and the Democratic contract down from 72 to 65, on roughly 300 contracts across the whole week. Then the quote screen moved further than the price history did. The resting 30-cent bid on the Republican contract was pulled, leaving a book of 6 bid against 34 asked, so the contract's midpoint reads 20 cents even though not a single contract has traded in the past 24 hours and the last print is still 34. That 20 is the artifact of a missing bid, not a price anyone paid. No public development lines up with any of it — no merits ruling in LULAC v. Abbott, no shift in the polling aggregates — so the honest description is a handful of small trades and one withdrawn order on a thin board, and this page will not dress that up as a story.
Every Seat's Number
| Seat | P(Democratic control), round 1 | Final (after revision) | Conditional hold priced in round 1 |
|---|---|---|---|
| Claude Fable | 62.5 | 64.0 | 74% |
| Claude Opus | 72.7 | 74.0 | 82% |
| Claude Sonnet | 68.5 | 68.0 | 80% |
| ChatGPT | 66.7 | 69.0 | 77% |
| Gemini Pro | 68.6 | 69.0 | 79% |
| GLM | 69.6 | 72.0 | 82% |
| Kimi | 77.0 | 76.0 | 85% |
| DeepSeek | 53.8 | 69.8 | 80% |
Every seat on this panel is graded against real market settlements — records to date: Claude Fable 82% on 576 graded calls · Claude Opus 83% on 646 graded calls · Claude Sonnet 82% on 624 graded calls · GPT 84% on 5,658 graded calls · Gemini 86% on 4,423 graded calls · GLM 81% on 2,437 graded calls · Kimi 82% on 2,414 graded calls · DeepSeek 80% on 2,462 graded calls. Recomputed daily; the full scoreboard is public.
More live boards from the same panel: House control 2026, where the same two-contract question settles first, has Democratic control at 84.5¢ · 2026 Senate control has Republican control at 52.5¢ on the same election night · the 2028 presidential election board scores all 30 listed names. The 2026 House and Senate prices were fetched August 27, 2026, 4:40 AM ET.
Seven of the eight discounted the hold, pricing it between 74% and 82%, under the six-of-seven rate, and only Kimi sat at 85%. The panel is that close to unanimous that the postwar carry-over rate was earned in an era of large majorities and does not survive contact with a chamber that currently seats 218 Republicans, 212 Democrats, one independent who caucuses with the Republicans, and four vacancies. The eight final answers, sorted, run 64, 68, 69, 69, 69.8, 72, 74 and 76; the last trade sits at 65 cents. When the panel ran on August 23 the market's midpoint and the panel's median were a tenth of a point apart; this week's drift puts the price history about four cents under the median, still inside the band the eight models span, and neither method treats the streak as being as safe as 18 straight elections make it look.
Where The Panel Changed Its Mind
Round two moved the board 16 points at the bottom and barely touched the top. The biggest revision came from the seat alone in the low 50s, which caught its own error: it had set aside extra probability for a messy, undecided outcome, then decided that hedge was not earned.
"I held my step-1 (83% D) and Democratic conditional-hold (80%) estimates but corrected a combination error that had overweighted divided-outcome tail risk. Peer arguments consistently applied the base rate without a large tail, and I adjusted Republican conditional hold from 85% to 80% to reflect narrow-majority vulnerability, aligning with the symmetric discount applied to Democrats." — DeepSeek
Two seats moved on how polling error is measured. The card's graded-poll archive shows presidential-year polls overstating Democrats by 3.7 points in 2016 and 4.7 in 2020, while the two midterms without Trump on the ballot were nearly unbiased, minus 0.3 in 2018 and plus 0.9 in 2022. Round one, both had applied the presidential-year haircut to the whole estimate.
"Moved 73 to 74 on one specific correction. In round 1 I let the Trump-era presidential-year poll bias (+3.7 in 2016, +4.7 in 2020) shade my estimate down as a general 'presidential years overstate Democrats' caution. Forecaster A's separation of the steps forced me to look at the archive again: the bias splits cleanly by whether Trump was on the ballot." — Claude Opus
A third seat caught itself double-counting the maps: the public 2026 forecast it leaned on was already built on the enacted lines, so its extra redistricting deduction charged the same toll twice.
"I was double-counting redistricting headwinds in Step 1: the fiftyplusone model published August 3, 2026 was already built on the enacted maps (Texas HB4 signed August 2025, Supreme Court reversal April 2026, all other redraws in place), so my 5-point haircut from 85% to 80% discounted for maps the model already absorbed." — GLM
And one seat sat at the center and held its ground, shading 68.5 to 68, which in a convergence round is its own kind of evidence.
"Held. The eight peers span 54-77% with a mean of 67.5 — my 68% already sits at the center, so this isn't a case of an isolated outlier needing correction. I checked each peer for evidence I hadn't priced: Forecaster C's move to 54% rests on an unexplained ad hoc 'tail for a divided outcome,' not new evidence — I don't adopt unjustified hedges." — Claude Sonnet
Round two barely moved the hold column: GLM raised its Democratic hold to 83%, Kimi held at 85%, DeepSeek held at 80%, and ChatGPT recorded that its 77% had been "too punitive relative to the 6-of-7 carry-over base rate."
Hottest Prediction Markets Right Now
- 2028 Democratic presidential nominee · $190M traded
- 2027 Pro Football Champion · $70M traded
- 2028 U.S. Presidential Election winner? · $62M traded
- 2028 Republican presidential nominee · $59M traded
- Pro Baseball Champion · $59M traded
- Bitcoin price at the end of 2026 · $32M traded
Every market above links to our full AI model verdict; browse them all on the OddsShopper prediction markets hub, and see how every settled call actually scored on the full graded scoreboard.
What Would Change The Panel's Mind
Both sides of this board are priced on a real case, and each gets the same room.
The case carried by the 65-cent side. Two gates both favor the same party right now: a midterm environment with approval more than 21 points under water and a generic ballot above six, then a record of 18 straight presidential elections without a flip and six of seven midterm flippers holding. The five seats Texas drew for itself are still being defended in court, while California's counter-map is written into state law through 2030.
The case carried by the 34-cent side. The 2026 election has not happened, a Democratic win can arrive at the bare minimum, and a 218-seat majority is a rounding error from gone. The redraws still net Republicans more seats, presidential-year polling in the Trump era overstated Democrats by three to five points, and 2028 will be an open race under the 22nd Amendment with two nominees nobody can name. And the contract settles on the party of the Speaker on February 1, 2029, not on election night, which in a chamber decided by a handful of seats is not a formality.
Every seat named its own falsifiers, and these four double as this page's re-score triggers.
- The November 3, 2026 Seat Count. The single biggest input, and it resolves in ten weeks. A Democratic majority at or above 230 seats pushes the conditional hold back toward the historical 86% and this number toward the mid-70s; 218 to 220 pushes it the other way, hard. One seat said 225 or fewer Democratic seats would move it toward a coin flip.
- A Merits Ruling In LULAC V. Abbott. A ruling that strikes the 2025 Texas lines before 2028 removes seats from the Republican column the current price assumes; a ruling that upholds them removes an upside tail the panel is crediting to Democrats.
- The Generic Ballot And Approval Through The Fall. A sustained move of two to three points in the generic ballot, or an approval recovery toward even, weakens the leg that carries most of the 69%. A rally-scale national event before November 2026 would re-open the 1998 and 2002 precedents.
- The 2028 Nominees. Coattails are the one mechanism that has ever flipped a House in a presidential year, and the two cases did not look alike: 1952 paired a 22-seat swing with Eisenhower's 10.85-point win, 1948 a 75-seat swing with Truman's 4.48-point upset. A race pointing at either shape is the specific thing that would justify a Republican number well above 34.
The Contract, The Book And The Timeline
| Venue | Kalshi |
| The Event | "2028 House winner" (Kalshi CONTROLH-2028), two mutually exclusive party contracts |
| The Contract | Per the settlement rules, verbatim: "If the Democratic party has won control of the House in 2028, then the market resolves to Yes" (mirror wording for the Republican contract) |
| How It Settles | Verbatim: "This market may be determined early based on a consensus of media calls projecting control of the U.S. House. See full rules for details. Otherwise, victory will be determined by the party identification of the Speaker of the House on February 1, 2029." |
| Next Catalyst | November 3, 2026: the midterm that decides which party defends the House in 2028 |
| Then | Merits proceedings in LULAC v. Abbott, which determine the Texas map for 2028; then the 2028 primary calendar and both parties' nominees, from spring into summer 2028 |
| Election Day | November 7, 2028; the new House organizes in January 2029; expected expiration February 1, 2029 (15:00 UTC) |
| Depth | About 10,800 contracts of lifetime volume across both sides, roughly 3,800 of them still open. The Democratic contract shows a five-cent gap between the best price to buy and the best price to sell, 65 bid against 70 asked; the Republican book is 28 cents wide, 6 bid against 34 asked, after a resting bid was pulled |
| Prices As Of | August 27, 2026, 4:40 AM ET (last trades, with each contract's live bid and ask shown on the board above) |
| Re-Score Policy | This page is re-scored on the same URL when the story moves, and every number on it is graded in public after settlement |
The depth row is the one to read twice. The 2026 House contracts on the same exchange have traded about 26 million contracts; the 2028 contracts, about 10,800, roughly 2,400 times fewer on a nearly identical question about the same chamber. A price with that little behind it is a real quote from a small number of participants, and the spread is part of the real cost on a position held for 29 months: five cents on the Democratic contract, 28 cents on the Republican one after this week's bid pull, against one cent on the 2026 board. This is the same thinness the page opened on, doing exactly what thin books do — the 65-cent last trade is the center of a band, not a decimal-precise reading of the crowd. Where these contracts sit legally is covered in our guide to prediction market legality; the same panel's work on the race that decides this board's first gate lives on the 2026 midterm board, and the day's full slate of graded calls sits on our daily Kalshi hub.
The Bottom Line
Traders and models still roughly agree on the 2028 House, at a 65-cent last trade against a 69.4% median, and both sit below what the postwar record alone would produce. Six of the eight models gave the same reason for that discount, that modern House majorities are too thin for a base rate built in the era of 250-seat chambers; one pointed at the redrawn maps instead, and one stayed on the historical rate. This week's seven-cent drift toward the Republican side arrived on about 300 contracts with no public development behind it, which is why it moves the board and not the argument. The number that actually resolves the argument arrives on November 3, 2026. If Democrats clear 230 seats, this board should drift toward the mid-70s. If they scrape in at 218, the streak that has held through 18 straight presidential elections gets its most serious test since Eisenhower. Either way this page gets re-scored on the same URL, and the panel's call gets graded in public.
Hero illustration: OddsShopper, in the house collage style. United States Capitol west front photo by Beethoven, licensed CC BY-SA 4.0; photos cropped, toned, and composited.
Prices on this page are Kalshi's book. If you also trade on Polymarket, code OS4 gets new users a $50 trading bonus on a $10 deposit — affiliate link; terms as stated by Polymarket; 18+, availability varies by state.
More on this: Kalshi 2026 House Odds: Which Party Wins The House? · House Control 2026 Odds: Kalshi's Ladder Tells on Itself · Kalshi 2026 Senate Odds: Which Party Wins The Senate? · How Many House Seats Will Democrats Win In 2026? Kalshi Vs. The AI Panel · TX-15 House 2026 Odds: De La Cruz Vs Pulido On Kalshi
