You can follow every race live on Manifold’s midterms dashboard.

The first race to watch: Kentucky’s 6th

One race to keep an eye on early in the night is Kentucky’s 6th District US House race. Only two states close polls at 6pm Eastern, Indiana and Kentucky, and it’s the most competitive race in either state based on the odds.

Traders currently give Zach Dembo (D) a 56% chance of victory over Ralph Alvarado (R) at 44%. This district voted for Trump by 15 points in 2024. They voted for their congressman Andy Barr by 26 points in 2024, however Barr is now running for the Senate seat, where he is an overwhelming favourite with a 95% chance of winning.

Kentucky’s 6th is open to voting for Democrats, though. Andy Beshear won here by 11 points in 2019 and 20 points in 2023 in the gubernatorial election, so the strong Republican lean of the district isn’t impossible to overcome for a Dem.

(Political Data Analyst Lakshya Jain)

There’s a conditional market on what happens if Dembo and the Democrats do win this district. Traders think Democrats will be in for an overwhelmingly good night if they can flip this Trump +15 district.

The big picture

Republican chances of holding the House have dropped to an all-time low of 2%. The Manifold dashboard shows Manifold traders have Democrats favoured in 242 seats.

Subscribe now

For the Senate, there’s a new market, Democrat+Independent Senate Seats after 2026 Midterms?, which will resolve on Associated Press race calls. That means you don’t have to worry about how an independent will caucus, if Fetterman switches parties, or having your mana tied up waiting until 2027 for a senator to be actually sworn in.

Democrats currently have a 92% chance to win the majority of governorships after the midterms. The last time they held the majority of them was in 2010.

Democrats currently hold 24 governorships and Republicans hold 26. They both have 18 of their governorships up for election. The Manifold dashboard has Dems favoured to win 22 of the 36 governor elections today, whilst Republicans are favoured in 14 of them. If both parties won all the ones Manifold traders favour them in, that would mean 28 governorships for Democrats and 22 for Republicans.

Where Manifold and Cook disagree

The Cook Political Report released their final House ratings yesterday, and they weren’t very brave in some races. They left California’s 22nd as a Tossup. Manifold has it as Safe D, with Randy Villegas at 94% to beat incumbent Republican David Valadao.

It’s a majority Hispanic district, and Hispanic voters are disapproving of Trump and Republicans in greater numbers. Valadao also only received 40% of the vote in the nonpartisan primary, and you’d assume most of the eliminated Democrats’ votes would go to Villegas. I think it’s fair to say Valadao is pretty much toast.

If you want to trade on who’ll win the 18 Cook Political Report tossup districts, there’s a market here, where the current favoured outcome is Democrats to win 12 or more, at 45%.

Before you live-trade the results

Be careful as the results come in. Here’s an example from Michigan in 2024.

From the 2024 Senate election in Michigan: Elissa Slotkin (D) is the blue line, and Mike Rogers (R) is the red. Slotkin won in the end by 0.34 points.

I found this tweet from Zachary Donnini of VoteHub interesting when comparing it to how the odds played out in 2024. About 4 hours after polls closed, the odds shifted in favour of Mike Rogers to 75%. That’s when VoteHub says you’d expect the most Republican-favourable votes to come in, before the strongly Democratic areas trickle in on Wednesday morning. You can see it in how Rogers’ odds steadily decreased the day after the 2024 election.

It’s a good reminder to be careful when live trading the results, and to try to have an understanding of where the votes are coming from. Zachary Donnini did a whole bunch of tweets on how different states may report their results, and I think they’re worth checking out on his Twitter account as a guide.

Subscribe now

Where to watch

Ryan McComb, a Manifold community member who works at VoteHub, will be on the CBS News/VoteHub election night livestream, so if you need a livestream to watch the results, make sure to check it out!

- Jack

On Sunday morning, a sitting Democratic congresswoman walked into a Detroit church beside the Republican Senate nominee and endorsed him. By Sunday night, the final New York Times/Siena polls were out, and Democrats led in all seven Senate races they surveyed, Michigan included.

Subscribe now

The final polls

The New York Times and Siena University released their final midterm polls on Sunday, taken Oct. 23 to Nov. 1. Democrats lead in every Senate race surveyed:

• Michigan: El-Sayed 53, Rogers 42

• Texas: Talarico 50, Paxton 43

• Maine: Jackson 51, Collins 45

• Ohio: Brown 49, Husted 44

• Kansas: Hamilton 49, Marshall 44

• Alaska: Peltola 48, Sullivan 46

• Iowa: Turek 48, Hinson 47

In 2024, NYT/Siena’s final polls overestimated Democrats by about 4 points on margin on average. Their final Texas poll had R+10, and the result was R+13.7. Even with a similar miss this year, Democrats would still win 5 of the 7 races. Alaska and Iowa would flip, but Michigan, Texas, Maine, Ohio, and Kansas would hold, some narrowly.

Republicans shouldn’t count on a polling error to save them, though. NYT/Siena updated their weighting model, and they say it would have been 1.6 points more accurate on their 2024 polls. In other good polling news for Dems, AtlasIntel, which Nate Silver rated the second-best pollster of 2024, just released a poll showing Talarico up 6 in Texas and Democrats winning the generic ballot by 14.1 points among adults.

Democrats won the 2018 House popular vote by 8.6 points and took 235 seats. Redistricting has made the map harder for them this time, but a D+14 environment would still push them well past 235. You can trade on how many House seats Democrats will win: currently they are 59% to win 240 or more.

Michigan

Traders give Abdul El-Sayed an 82% chance of winning. The NYT/Siena poll was taken before Rep. Haley Stevens, who narrowly lost the August primary to El-Sayed, endorsed Mike Rogers on Sunday. El-Sayed dropped from about 80% to 70% after the endorsement, then rebounded to 82%, likely on the good poll and a wave of Democratic support in response to Stevens’ endorsement, including what his campaign says was $1.2M raised in 4 hours.

Rep. Kristen McDonald Rivet, the only Michigan House Democrat in a Trump-won district, now says she’s voting for El-Sayed and urges others to. She hadn’t backed him specifically before. Other House Democrats are angry: Rep. Marc Veasey of Texas, who says he was the first House Democrat to hold a fundraiser for Stevens’ Senate campaign, says he’d be the first to make a motion to remove her from the House Democratic Caucus, which has a rule against campaigning for non-Democrats in federal elections.

There’s little time left for the endorsement to change minds, though. As of Oct. 30, 1.86m people had voted early, compared with 1.52m at the same point in 2022, a 22% increase. With 4.4m votes cast in the 2022 governor’s race, that’s already over 40% of the 2022 total.

One factor getting little attention is religion. In an August Michigan poll from Michigan State University, 31% said they were much less likely to vote for a Muslim candidate and 7% somewhat less likely. That’s a hit before anyone even looks at his policies, though plenty of those voters probably weren’t going to vote for a Democrat anyway.

It also means that underperforming Benson, the Democratic nominee for governor, doesn’t automatically make El-Sayed a bad candidate or too far left. In 2018, incumbent Debbie Stabenow, a standard Democrat, ran 3.1 points behind Gretchen Whitmer in margin on the same ballot. Benson and other more moderate candidates also haven’t had to deal with a primary opponent who supports their opponent.

I’ve created a market on how El-Sayed will perform relative to Benson. You can also trade El-Sayed’s winning margin, but traders aren’t buying the NYT poll just yet: there’s currently only an 8% chance he wins by 11 points or more.

The other Senate races

The poll moved the odds, with some sharp shifts shortly after release.

• Maine: Troy Jackson’s (D) chances rose from 68% to 73%. Susan Collins is concerned.

• Kansas: Democrats haven’t won a Senate seat here since 1932, but Adam Hamilton’s chances rose from 50% to 61%. His main attack on incumbent Roger Marshall (R) is that Marshall, an obstetrician, sued former patients over medical debt. Marshall released a video defending it as common practice he did to keep the lights on, while often providing free care. According to the NYT, his own team advised him against releasing it. About 750 Kansas medical professionals signed a letter criticizing his debt collection, and a class-action lawsuit filed Oct. 29 alleges his practice’s 18% interest rate broke Kansas’s 15% legal cap. Marshall’s chances have fallen from 90% to 39% since the start of the year.

• Texas: James Talarico (D) rose from 70% to 75% on the poll and strong early voting data for Democrats. Ted Cruz has said Republicans could lose every statewide office in Texas, that Republicans shouldn’t think the polling is all wrong, and that Democrats are “massively out-turning out” Republicans.

• Ohio: Sherrod Brown (D) has a 76% chance. He lost in 2024 as an incumbent but hopes to return to the Senate now.

• Alaska: Mary Peltola (D) is at 78%, about where she was before the NYT Dan Sullivan story dropped.

• Iowa: Still close, with Josh Turek (D) at 46%.

After the midterms

I’ve created two markets on how new Democratic senators will vote next year:

• Which senators will vote as left-wing as Elizabeth Warren? El-Sayed, Peggy Flanagan (Minnesota) and Troy Jackson are the current favorites, with Angie Nixon level with Jackson at 70%.

• Which senators will vote for at least one Trump judicial nominee in 2027? Of the 47 current Democratic senators, 20 have voted for a Trump judicial nominee since he returned to office.

It will be interesting to see which newcomers resist Trump and which help keep confirmations moving. If Democrats have a good night, some of these senators will represent states Trump won by 10 or more, and most won’t face voters for six years. Do they spend two years resisting Trump and turn bipartisan once he’s gone, or confirm judges now and risk upsetting the Democratic base?

The ‘26 Midterm Contest

The Michigan margin market is one of 40+ in the contest. When they all resolve, the top 12 traders on the leaderboard win mana prizes, with 6k mana for first. Examples of markets:

• Will Ossoff and Roy Cooper’s combined margin be 25 points or more?

• Will Denver and DC both pass foie gras bans?

• How quickly will the Associated Press call the Georgia Senate election?

• Will Republican chances to win the Senate go back above 50%?

Check out the rest of the markets here.

Also on Manifold

There’s a new perpetual market tracking inflation plus unemployment, built on Adjacent Markets’ Misery Implied Index (market-implied U-3 unemployment plus market-implied CPI-U inflation).

Adjacent also runs political indexes, including a BLUE Index for Democrats’ 2026 and 2028 chances, which has kept rising with their polling, and a RED Index for Republicans.

Check out the Manifold market on the BLUE index here.

- Jack

Democrats lead the generic congressional ballot by 6 points in the latest YouGov/Economist poll (October 23-26, 2026), but by 11 points among likely voters. The gap largely reflects enthusiasm: 88% of Democrats polled were classified as likely voters, whilst only 81% of Republicans were.

The Trump coalition includes more low-propensity voters who show up only in presidential years, along with some who are disappointed in the administration. In the October 2026 poll, 95% of Harris 2024 voters say they made the right choice in 2024, compared with only 86% of Trump 2024 voters. If Republicans can bring enough of them back, they could hold some House seats in tight races, but keeping control of the House looks very unlikely. Traders give Republicans only a 4% chance.

One of the biggest factors hurting Republicans is the economy. The October 2026 YouGov poll shows 59% of voters think the economy is getting worse. That’s a big contrast to the 2018 midterms, when a late October YouGov/Economist poll found only 23% of voters thought so. In 2018, Democrats won 235 House seats and the House popular vote by 8.6 points. Given opinions on Trump and the economy, this year’s margin could be even bigger.

The direction of the country question tells a similar story. Before the 2018 midterms, 49% of voters said the country was on the wrong track. In a February 2025 YouGov/Economist poll after Trump had become President again, it was 50%. In the latest October poll it’s 72%.

Strength of feeling about Trump has also shifted. In February 2025, 30% of voters strongly approved of him and 37% strongly disapproved. Now 22% strongly approve and 50% strongly disapprove, a swing from -7 to -28.

One constant reminder voters get is gas prices, which currently average $4.26 nationwide, up from $3.12 on Trump’s inauguration day. Traders expect it to stay at roughly that level on Election Day.

The Senate

Republican chances in the Senate have fallen sharply this year. Traders now give them only a 26% chance of winning 50 or more seats (enough for control with the VP tiebreaker) in Tuesday’s election.

Iowa is one of two Senate races Manifold currently rates a toss-up on the election dashboard. Traders give Republican Ashley Hinson a 57% chance of beating Democrat Josh Turek, even though the polls are essentially tied: the average across five aggregators is 46.0% each, with individual aggregators ranging from Hinson +0.8 to Turek +1.2. Iowa is also the favourite, at 18%, in the market for which Senate race will have the smallest margin of victory.

Democrats currently hold 47 Senate seats, including independents who caucus with them. They need 51 for a majority, since Republicans hold the VP tiebreaker at 50-50. On Manifold, 52 seats is the most likely outcome, but at just 14% it is far from certain. A 52-seat result would give Democrats a one-seat buffer if Fetterman or another senator left the party or refused to back a particular bill.

Alaska

In the Alaska Senate race, traders give incumbent Republican Dan Sullivan only a 32% chance of holding his seat against Democrat Mary Peltola.

The New York Times reported on Wednesday that aides to Peltola’s campaign and staffers at the Democratic Senatorial Campaign Committee worked to encourage spoiler candidates to enter the race, to try to take votes from the incumbent Republican. According to the Times, Peltola’s former campaign manager said Peltola wasn’t directly involved but was aware of it. Peltola’s campaign and the DSCC deny any involvement, and the former campaign manager has since tried to retract her statements to the Times.

One of those candidates is also named Dan Sullivan. He filed to run as a Republican. Election officials viewed that as bad faith and tried to remove him from the ballot. After the Alaska Supreme Court put him back on, they listed him under his registered voting name, Daniel J. Sullivan Jr., with no party label to avoid confusion with the incumbent, who is marked “(Incumbent)” on the ballot. Records show the spoiler Sullivan donated to Peltola’s House campaigns in 2022 and 2024 and to an Illinois left-wing House candidate in 2025, which raises doubts about his attempt to file as a Republican.

How the candidates appear on the ballot. Republican Gerald Heikes is also running.

Sullivan Jr. took 2.5% and finished third in the open primary, and traders give him an 87% chance of finishing third again. On Truth Social, Trump has called the ballot a “TOTAL DEMOCRAT SCAM” meant to confuse voters.

Early voting is strong: 58,168 votes had been cast as of Tuesday afternoon, October 27, about 22% of the 267,047 cast in the 2022 midterms. All of those votes were in before the Times story came out.

In the Alaska governor’s race, Republican Senator Lisa Murkowski has crossed party lines to endorse Democrat Jonathan Kreiss-Tomkins, who traders give a 90% chance of winning. Republicans are hurt by the open primary and ranked-choice voting. One Democrat and three Republicans advanced from the primary. Trump endorsed Bernadette Wilson, but the outgoing governor hasn’t endorsed anyone and has told voters to rank all three Republicans. Ballots that don’t rank all three could be exhausted, and a split party fighting itself makes it hard for Republicans to win.

Trump recently called Murkowski the worst Republican senator by far and a disgrace to her state, and said she was only reelected in 2022 because of “corrupt” ranked-choice voting. Traders give her a 32% chance of leaving the Republican Party before 2029.

Alaskans are also voting on a measure to repeal ranked-choice voting and restore party primaries. Trump has endorsed it, and Elon Musk is one of its funders, but traders give it only a 33% chance of passing. A very similar measure failed in 2024 by 743 votes (0.23%). If it passes, Murkowski would be vulnerable to a primary challenge in 2028, because she relies on crossover Democratic votes. She won in 2010 as a write-in candidate, and in 2022 she picked up about 20k votes from the eliminated Democrat, against about 2k for her Republican opponent.

So on Tuesday, watch Alaska twice: once for which Dan Sullivan voters pick, and once for whether they keep the system that made that question possible.

All of it is tracked on Manifold’s midterms dashboard.

How the dashboard works

This midterm cycle, Manifold has markets for all 35 Senate and 36 governor races, plus 421 of the 435 House races (the other 14 have only one party on the ballot), and 124 statewide ballot measures.

For the House, Senate, and Governor races, the dashboard rates each race as Safe, Likely, Lean, or Toss Up for each party based on the current market odds.

Safe means traders give that party or candidate a 90% or greater chance of winning, Likely is 75–89%, Lean is 60–74%, and Toss Up covers races where no party is over 60%.

The House

At the time of writing, Manifold traders rate 211 seats Safe for Democrats and 141 Safe for Republicans. That leaves 83 races in the more competitive Toss Up, Lean, and Likely categories.

The 83 competitive House seats on Manifold’s midterms dashboard, as of October 25.

The odds and ratings update after every trade, so check out the map yourself to get the most up-to-date ratings and to make your own predictions. The ratings are also only a snapshot in time: ‘safe’ races can easily become competitive for a number of reasons, and some races will shift the other way, from competitive to safe.

One example of a shift is in Maryland’s 1st district, where Andy Harris, the incumbent Republican who won by 22 points in 2024, has dropped from 92% to 84% in the last two weeks, shifting the race from Safe R to Likely R. The shift is probably due to a poor debate performance that went viral, where he blamed rising diesel prices on Democrats’ environmental regulations and got very mad when the crowd laughed at his response.

In a post-debate press conference, he got mad at a student journalist who was asking questions. He told her to stop recording, citing Maryland’s wiretapping law, and threatened to take her camera for recording him without his consent, all this despite there being no expectation of privacy in a press conference!

Right now, there are 6 House races rated as Toss Up by Manifold traders. One of them is Maine’s 2nd Congressional District.

Maine’s 2nd District

Former Republican governor Paul LePage is running against Democrat Matt Dunlap. There is no incumbent after Jared Golden (D) decided not to seek reelection.

LePage is slightly favoured by traders at 56% to Dunlap’s 44%.

If elected, LePage would be one of the oldest first-term congressmen ever, at 78.

LePage served as Governor from 2011 to 2019. He ran for a third, non-consecutive term in 2022, against incumbent Janet Mills. He lost the election statewide by 13.2 points, but he won the 2nd congressional district by 2.4 points.

The political environment looks much worse for Republicans this year, though, given President Trump’s unpopularity. He has a net approval of -13 in this district according to a recent poll, despite having won this district by 10 points in 2016, 7 in 2020, and 9 in 2024.

LePage is campaigning on being close to Trump. He has broken with him slightly on the Iran War, tariffs, and even ICE, after it was reported that ICE was trying to deport his son’s father-in-law. But LePage believes his biggest selling point is the ability to get into the White House and meet with Trump, arguing that his opponent won’t be able to get meetings like this.

Matt Dunlap is the current Auditor of Maine, and previously served as Secretary of State from 2005 to 2011 and 2013 to 2021. Both positions are appointed by the state legislature in Maine.

Dunlap is a progressive who supports Medicare for All, and originally entered the race to primary challenge moderate Democrat Jared Golden, before Golden announced his retirement. Likely due to fears he was too left-wing for the district, Democrats in DC originally opposed Dunlap’s candidacy, endorsing a primary opponent, whom Dunlap beat in the final round of ranked-choice voting after placing second in the first round. Since the primary, they have gotten behind him, including spending $2M in just the last month to help him win.

One factor hurting Republicans in Maine is the price of heating oil, which approximately 50% of Maine households rely on. A gallon of heating oil in Maine averaged $6.04 at the start of this week, compared to $3.30 on October 20 last year. Heading into winter, this will be on a lot of people’s minds. Running strongly against the Iran War should help Dunlap and Troy Jackson, the Democratic nominee for Senate, on this issue. After lobbying from Maine’s senators, Trump released heating oil from the reserve this week to try to help.

Collins’ reelection odds on Manifold, as of October 25.

The Senate: Maine

The tough environment in Maine hurts Susan Collins. Her odds fell to 33% last week before recovering to 38%, and traders still have her as the underdog.

A poll for Fox News, released on October 22, has the race tied amongst likely voters at 49% each for Troy Jackson and Susan Collins. Amongst registered voters, Jackson leads 50% to 48%. The same poll found that 46% of respondents were concerned Jackson is too liberal, while 53% think Collins votes with Trump too often.

The polling averages show a small but consistent lead for Jackson.

Maine Senate polling averages, compiled by Wikipedia, updated October 16.

Governor: Georgia

Georgia governor’s race on Manifold over the past week.

The only current Toss Up governor’s race is Georgia. In the last week, the Democrats’ chances have risen from 48% to 59% according to Manifold traders, with potentially positive early vote data explaining this. The Republican nominee, Rick Jackson, has said he’s concerned about low turnout in rural counties, which are deeply Republican. Statewide in-person early vote numbers were up on day one of early voting, with 40% more votes cast compared to the 2022 midterms, and more than double the 2018 midterms.

However, Rick Jackson still leads in all the polling averages.

Georgia governor polling averages, compiled by Wikipedia, updated October 21.

Rick Jackson is a billionaire who made his wealth as a founder of a healthcare company. He has spent nearly $200m on his campaign, which is his first run for public office. He beat the Trump-endorsed Lt. Governor in the Republican primary runoff, after coming second in the first round. Jackson is running as a conservative outsider who has the business skills to deliver as Governor.

Keisha Lance Bottoms, the Democratic candidate, is a former mayor of Atlanta and former senior advisor to President Joe Biden. Biden made one of his only political endorsements since leaving office for Bottoms, filming a video saying she’s “something special” and was “invaluable to the administration.”

Bottoms would be the first Black governor and first woman governor of Georgia, the first Black woman governor in American history, and the first Democratic governor of Georgia since 2003. Bottoms has standard liberal policies and has been casting Jackson as too close to Trump. She has been holding rallies with Jon Ossoff and will be hoping his likely big win (he currently trades at 70% to win by 8% or more) can help her candidacy.

A Bottoms win could be a big factor in a potential Ossoff presidential run, as having a Democratic governor ensures a Democrat would be appointed to the seat if he leaves the Senate. Traders currently give Ossoff a 24% chance of not finishing his term if he’s reelected.

Every one of these races is live on Manifold’s midterms dashboard. With less than two weeks to go, check back as early votes come in, or make your own predictions.

- Jack

Perhaps the most candid account of the Republican midterms campaign so far comes from that leaked recording. Someone at the fundraiser had asked Paxton (Trump’s handpicked Senate nominee in Texas) what was dragging his numbers. When another …

Read more

It could help explain why, at the time of writing, Manifold traders give Democrats a 60% chance of winning a Senate seat in a state where they haven’t won statewide since 1994. Why Republicans’ new House map, drawn to take five seats from Democrats, looks likely to net them only two. And why Greg Abbott, the longest-serving governor in the country, is heading into what looks like the closest race of his career.

You can see how all of these races are moving on Manifold’s 2026 midterms map.

The Senate: Talarico vs. Paxton

Texas Senate on Manifold’s midterms map, as of October 16. Party and candidate markets are trading within a point of each other.

Democratic nominee James Talarico’s odds are the highest they’ve been all race. He was trading around 30 to 40% in the months after the primary and has been climbing ever since.

Democrats thought they had a great chance with candidate Beto O’Rourke in 2018, but it’s lining up to be even better for Talarico in 2026. He currently leads in the polls, something O’Rourke never achieved.

At this point in the election in 2018, O’Rourke was down in the RealClearPolitics polling average by 7%. Now, Talarico leads by about 2.8%.

O’Rourke went on to lose narrowly to Republican Ted Cruz, 48.33% to 50.89%.

Polls from RealClearPolitics

But the polling average isn’t always right in Texas. In both 2020 and 2024, the RealClearPolitics polling average underestimated Donald Trump’s margin by about 4.5%. In 2018’s Cruz-O’Rourke race, the polling average actually overestimated Cruz by 4.2%. A miss of that size in either direction would wipe out Talarico’s lead, or double it.

Democrats haven’t won a Senate race in Texas since 1988, part of the longest statewide losing streak in the country. And just two years ago, Trump carried the state by nearly 14 points.

Right now, traders give Talarico an 83% chance of outperforming O’Rourke, who ended up losing his race by less than 3%.

Ken Paxton

Paxton is a weak candidate: in 2018 and 2022, he had the lowest percentage of the vote total among all Republican statewide candidates.

*Average of six other Republican statewide candidates: Governor, Lt. Governor, Comptroller, Land, Agriculture and Railroad Commissioner. Source: Texas SOS.

Paxton has also faced many scandals, including cheating on his wife, State Senator Angela Paxton, who filed for divorce in July 2025 on biblical grounds. Despite this, she recently recorded an ad for a Super PAC in support of his campaign, urging people to vote for the “Republican ticket,” while notably avoiding mentioning him by name.

Other transgressions include a 2015 indictment for securities fraud, allegedly stealing a $1,000 pen, and allegedly abusing his AG office to help a friend target business rivals. Paxton added a new scandal to the list this week when the Wall Street Journal reported on a video of him celebrating New Year’s Eve 2024 in Dubai with his girlfriend and two men who were then federal fugitives, one of whom had made a sizeable donation to his campaign.

It seems unlikely Paxton will debate Talarico; traders give it a 21% chance. Talarico has been pushing for one for months, even challenging Paxton to a debate live on Republican-friendly Fox News. Paxton still hasn’t agreed, even after Trump posted on Truth Social last week, calling Talarico “Talafreako” and calling Paxton “ready, willing and able.”

It was announced this week that the Stark Freedom PAC plans to spend up to $2.5m to boost third-party candidates in 8 races nationwide, including the Texas Senate race. Among these candidates is Libertarian Ted Brown, who traders currently say has a 52% chance of getting 2% or more of the vote.

The gerrymander that backfired

Texas House odds by district. The blue in South Texas is where the GOP’s redraw targeted Democratic seats.

For US House races, Texas is using a new map that state Republicans drew to flip five Democratic-held seats: the 9th, 28th, 32nd, 34th and 35th.

However, Democrats are favored to hold 3 of these 5 seats: the 28th (92%), the 34th (83%), and the 35th (61%).

Democrats have two things working in their favor: good candidates, and a map that may have been drawn on bad assumptions.

For example, in 2024, Trump would have won the new Texas 28th district by 10.4%, while he won the old 28th by 7%. However, despite growing more Republican at the presidential level, the district has become more Democratic down-ballot: under the new lines, it would have voted 53.9% Democratic in 2024’s House races, up from 52.7% in the old district.

Republicans have assumed that the gains Trump made in 2024 with Hispanic voters will transfer over to other Republicans, but it’s not working out that way.

All five targeted districts are majority Hispanic, and a big reason the redistricting isn’t working is the shift in the vote toward Democrats among Hispanic voters. The polling in the Texas Senate race shows this. In 2024, YouGov and the Rose Institute had a poll showing Colin Allred, the 2024 Democratic nominee for Senate, winning over Hispanic voters by 4%; in 2026, the most recent YouGov poll shows Talarico winning Hispanics by 27%.

Democrats who specialize in Hispanic voter outreach say economic issues drove Hispanics away from Democrats in 2024, and that the same issues are driving them away from Republicans now. These voters are a swing group and shouldn’t be expected to vote the same way every election or follow a party line.

Trump was seen as running on a populist economic message, but that doesn’t guarantee another Republican has the same appeal. You can see this in how Ted Cruz ran behind Trump in some of these districts, like the new Texas 28th, where Cruz would have only won by 0.2% compared to Trump’s 10.4%.

District by district

TX-28 (Democrats 92%). Incumbent Henry Cuellar is a strong candidate for the area despite corruption scandals and being pardoned by Trump (who called him ungrateful when he didn’t switch parties). In 2024, he won his House race by nearly 6% despite Kamala Harris losing the district by 7%. He ran double digits ahead of Biden in 2020 and Hillary Clinton in 2016. Republicans have basically given up on defeating him this time, with the House Republican Super PAC pulling millions of dollars in advertising and sending the money elsewhere.

TX-34 (Democrats 83%). Incumbent Vicente Gonzalez is another overperformer, winning by 3% in 2024 as Harris lost his district by 5%. He is a conservative Democrat, one of only 7 in the party who voted to fund DHS 2 weeks after the shooting of Renee Good. The redrawn district is 77% Hispanic and would have voted for Trump by 10%.

TX-35 (Democrats 61%). This district would have voted for Trump by 10.4% in 2024.

TX-9 (Democrats 18%). There is no incumbent here. A recent poll shows the Republican candidate leading 48% to 42%; however, this is a district Trump would have won by 20%.

TX-32 (Democrats 12%). This district would have voted for Trump by 17.7% in 2024.

Keep an eye on these two celebrity races

There are two interesting and competitive races that aren’t in districts Republicans targeted for redistricting, but each has a candidate with a big following outside of politics.

TX-15 (Democrats 82%). This district voted for Trump by 18% in 2024, but Democrats have a star candidate in Latin Grammy Award-winning Tejano music artist Bobby Pulido. He has 10.4 million monthly listeners on Spotify, and a recent poll has him up 50% to 45% over incumbent Republican Monica De La Cruz.

TX-23 (Democrats 32%). Republicans have their own celebrity here: Brandon Herrera, also known as “The AK Guy.” Herrera’s YouTube channel focuses on a variety of content about firearms and has 4.3 million subscribers. Traders favor him, and Trump won the district by 15%, but this is another majority-Hispanic district where the swing away from Republicans could hurt his chances. Herrera also has a history of making vulgar comments about women, which House Speaker Mike Johnson says he has apologized for.

Even Abbott looks nervous

In the Texas governor’s election, it’s incumbent Republican Governor Greg Abbott against Democratic State Representative Gina Hinojosa. Traders give Abbott a 76% chance.

Abbott, first elected in 2014, is looking at his closest race yet, and he’s showing signs of being at least a little bit concerned. In November last year, he stood with the CEO of Google and said Texas is the epicenter of AI development; now he’s paused new permits for data centers.

He won his first term by 20 points, his second by 13, and his third, against O’Rourke in 2022, by 11. Currently, he leads the RealClearPolitics polling average by 2.2%.

What to watch on election night

Keep an eye on the majority-Hispanic districts: the 28th, 34th, 35th, 15th, and 23rd. If Democrats hold or flip them, the Hispanic swing in the polls is real, and Talarico’s 60% probably isn’t too high.

Manifold’s election page tracks every Senate, House, and governor race in real time, and you can trade on how you think Texas will go.

-Jack

The most useful thing about prediction markets may not be learning to anticipate future outcomes, but learning to think more carefully about the present and notice how you interpret the world.

This is a long one; you can cut straight to the punch if you must…

Asking better questions

Curiosity becomes operational

I guess there are totally incurious people in the world, but I haven’t met many of them, and I doubt they’re three paragraphs into this article, so forgive me for speaking in absolutes here:

We’re all curious about the world! To varying degrees, we all have questions, and some of us wonder not only what is or will be, but what others think of that, too.

But being curious isn’t the same as knowing what it is you’re asking. Sound silly? Trust me, you’d be surprised how common it is to not really know what you’re asking until you’re forced to write resolution criteria for it.

Here’s a little example: Will a new Lyme disease vaccine be available by the end of 2027?

Seems like a simple enough question, right? In face-to-face conversations, it’s probably sufficient. After all, if you’re having a discussion about it, you’ll learn along the way what the person you’re chatting with counts as “available.”

But what happens when you need to write clear and objective criteria, and you can’t just go with the flow when people come along with different ideas? Chatting about it with the guy sitting next to you at your airport gate when you see a poorly captioned CNN segment on a muted TV is much different than needing a reliable framework for what qualifies to resolve a market around the question you mean to ask.

Here, “available” needs to be much more concrete.

Do you mean

• Approved by the FDA?

• Approved anywhere? Or in the country where you live?

• Actually launched commercially?

• Prescribed outside of a clinical trial?

• Stocked somewhere and technically obtainable by a patient?

• Routinely available through ordinary doctors’ offices or pharmacies?

• Covered by insurance?

• Featured in an RFK Jr. podcast episode about why it’s evil?

What happens if it’s technically available, but so highly constrained that it won’t be broadly accessible until much later? Will it count if it’s only available to certain ages or risk groups? Interrogating the question you initially asked may reveal that it’s not actually the question you want answered.

Of course, even if you try to be explicit, you may end up with some gaps. Thankfully(?), the comment section will often let you know what you’re missing.

Language is less obvious than you might think

There’s what you know, what you know you don’t know, and what you’re absolutely sure you know until you meet someone who’s equally certain of a wildly different interpretation.

That third category has been a rite of passage for many people who start writing markets on Manifold.

Let’s try one of my favourite semantic examples:

If I ask you, “Will Taylor Swift be photographed with Pope Leo XIV by January 1, 2027?” do you think I mean:

Will Taylor Swift be photographed with Pope Leo XIV on or before December 31, 2026 at 23:59?
or
Will Taylor Swift be photographed with Pope Leo XIV on or before January 1, 2027 at 23:59?

If you chose the former, congratulations! You’re right! Here’s an example of a thing you know!

Except, if you went with the latter, you’re also right! Oh no! It’s a trap!

I’m on Team December 31st, mostly because I’ve considered “by” to be synonymous with “before,” and it didn’t occur to me that it could mean anything other than that until I started writing questions or betting in other people’s markets.

This is easily the most common mix-up because, no matter whether one is more typical in common parlance, both options are defensible interpretations. What an annoyance of the English language.

There’s nothing that will teach you about the intricacies of language like people whose money (or mana) is riding, in part, on your ability to be clear in your question and criteria.

Theory of mind

So, what does it take to be clear in your question-writing? I could argue a great many things, like subject-matter knowledge, a firm grasp on what exactly you’re asking, and a solid lexicon for putting those ideas down into words. And those are all crucial skills, but the S-tier skill secret is theory of mind.

Now, this is one of my (many) favourite soapboxes, but that’s a rant valuable monologue better saved for another time and another Substack. Instead, I just want to highlight how far ahead you often are in both writing and predicting when you develop a good theory of mind.

Communication is less about what you intend to say than what you want another person to receive. The more time you spend in prediction markets (especially user-created ones!), the more you build a model of how people see and communicate things. As with the above on the silliness of language, you start to anticipate grey areas by putting your default aside and auditing from someone else’s point of view.

Instead of assuming a counterparty or market creator is a fool, you can put yourself in their mindset and ask what interpretation would make their question or bet make sense.

Once you build a habit of interrogating your own questions, you begin hearing other people’s questions differently, too. You become less inclined to answer at first read, and more inclined to ask what information the person is trying to get at. What are they trying to determine?

That habit follows you out of the market, and you begin listening not just for the literal question but for the thing the person is trying to point to.

Subscribe now

Learning how certain you actually are

Certainty becomes proportional

In casual conversation, we tend to speak in binaries:

• I think this will happen.

• I don’t think it will.

Or,

• There’s no way in hell this doesn’t happen.

• You’re out of your mind to think that’s remotely possible.

Wanna shake on it?

Maybe you make a handshake bet on who buys the next round, but ultimately, this structure demands there’s a winner and a loser.

A market asks a different question: How sure are you?

Proportionality matters. You may land on a market asking a question that you believe will resolve Yes, but how strongly do you believe it? It’s sitting at 55%; do you think it’s actually closer to 70%? Or 30%?

You start noticing how often “I think [xyz]” is doing a lot of work for beliefs held at very different levels of confidence.

A belief at 55% is fundamentally different than a belief at 90%, and participating repeatedly teaches you how to notice that difference internally. It changes certainty from something you feel into something you have to inspect.

Having something at stake changes your relationship to your own conviction

There are real-money sites out there, of course, and if you’re reading this Substack (and you’ve read this far), I’m going to keep moving along because I trust you know who they are. Manifold runs on play money, but that doesn’t mean you don’t have skin in the game.

Even with play money, there is a consequence. You have a visible position. You have a track record. Other people can see what you believed, and how strongly you believed it. If you burn through it, you have less mana to play with.

Suddenly, you have to ask:

• How much of my belief is based on information?

• How much is intuition?

• How much confidence does the evidence actually justify?

• What evidence would move me from 60% to 80%?

• What evidence would make me substantially less confident?

tl;dr: Am I going on vibes?

Prediction markets don’t only teach you to assign confidence. They train you to develop a relationship with your confidence, and notice what actually moves it.

Markets give you feedback about whether your preferred way of knowing actually works

You may believe you’re good at going on vibes, spotting political shifts, reading people, parsing scientific developments, anticipating outcomes, etc. And maybe you are! Maybe you’re even unusually good at it! But, how confident are you? how do you know?

I’m not saying there are some memory biases at play here, but I’m not not saying it. Without a ledger, it’s easy to kid yourself a bit, and after all, why not? It’s good to feel good about yourself!

Life may not come with a formal audit trail, but markets leave receipts.

Over time, you build a record of not only which calls were right, but how confident you were when you made them. Maybe your political instincts actually are unusually good! Maybe the “vibes” you were getting right are just more memorable than the ones you misinterpreted.

You’re not only on record to yourself, but in front of your counterparties, and you may find the evidence humbling. You’re reality-testing your own self-conception; do with that knowledge what you will.

Learning your relationship to risk

Markets reveal that risk tolerance isn’t necessarily a single personality trait

Over the years I’ve known many, many people who think I’m either very brave, very irresponsible, or some mix of the two. Within about three months, I went from making a joke about leaving the US for Australia to quitting my job, selling all of my stuff, booking a hotel for 3 nights, and flying one-way to a country I didn’t know anything about with two suitcases in hand. I don’t think I had time to think of it as brave, but I just assumed I’d figure it out.

Since then, I’ve been to dozens of countries on one-way tickets, traveled extensively as a solo female without a plan, moved to a couple of other countries, quit my job to backpack until I ran out of money more than once (I call these chapters “practice retirement”) and, perhaps most bravely, eaten street food all across SE Asia and India.

And yet, when I joined Manifold, I was fascinated to see how conservative I was with my bets. I was making little baby bets; I was anxious about outcomes; I found it particularly unsettling to lose mana. I suppose I’d been reserving all of my risk-aversion for the markets.

Prediction markets can highlight that “risk-taker” and “risk-averse” aren’t particularly useful global descriptions.

People have very different tolerances for:

• financial risk

• career risk

• uncertainty

• reputational risk

• social risk

• physical risk

and markets can bring those differences to light because they repeatedly force you to decide not only what you believe, but how much you’re willing to put on the line for it.

Markets create material for self-audit

Your trading history not only provides valuable information on how often you were right, but also produces data about your own decision-making. Over time, you can start asking:

• Am I chronically too conservative?

• Do I only make bold bets in domains where I feel personally knowledgeable?

• Do I hesitate even when I have an informal advantage?

• Am I willing to take low-probability/high-upside positions?

• When I’m wrong, is it because my evidence is bad, my probability estimate was bad, or my willingness to act on it was bad?

Over time, your portfolio not only gives you information about the world, but information about yourself as an instrument to interpret the world. That kind of self-knowledge is useful; you’re learning more about how you respond to uncertainty, loss, conviction, and risk. You wind up with a distinction between what you believe and what you’re willing to do with it.

Separating prediction from preference

A bet is not an endorsement

Okay, finally. Now’s my chance to Take Back a pithy statement often used in disingenuous ways:

Facts (and markets) don’t care about your feelings!

(If that saying doesn’t sound familiar, I pat you on the head, my sweet summer child. Don’t worry about it, just know I’m out here fighting the good fight.)

This is one of those things that becomes obvious once you’ve been around prediction markets for a while, but it can be surprisingly unintuitive. Someone buys Yes on an outcome you find awful, and the novice instinct can be to read that as support for the outcome rather than a prediction about reality. Or to be disgusted that a market even exists for what you consider a distasteful question.

But prediction markets aren’t a chance to vote with your wallet; they’re a mechanism for pointing to the truth.

What do I think will happen?
is not the same as
What do I want to happen?

And,

What do I think is true?
is not the same as
What would I prefer to be true?

If you can’t separate your desires from reality, you’re going to have to be comfortable just doing a token victory lap for the cause, and you might as well just leave your purse on a park bench while you do it.

Markets punish motivated reasoning in a very direct way

People naturally tend toward evidence that favours their desired outcome. Hell, there’s an entire Catalogue of Biases to explain this tendency through a ton of different lenses. You may think you can be pretty objective, but prediction markets have a way of making your biases more visible.

You overweight ambiguous information in your preferred direction, filling in the gaps yourself while discounting unpleasant evidence. You treat a prediction as a moral statement, and your aversion to aligning with (or profit-seeking from) the idea causes you to burn currency in protest without even noticing.

Once you stake something on the outcome, even play money and your reputation, wishful thinking becomes more expensive. Part of the training is learning to say “I hate this outcome, and I think it’s 70% likely.”

The ability to separate your values from your beliefs is an important civic/intellectual skill, but once you’re holding both in your soft little hands, where do you go from here?

Well, to start, you may find yourself watching the Super Bowl or the Oscars and feeling really awkwardly conflicted in a way that feels almost darkly comedic.

More usefully, once you stop allowing “I want this to be true” to serve as evidence, you have to get more serious about what does, and who you trust to provide it.

Learning whom and what to trust

Prediction markets teach source discernment rather than source allegiance

Author Michael Crichton coined the term Gell-Mann amnesia, noting that we often notice how badly media gets subjects we know about wrong, then turn the page and resume trusting its reporting on subjects we know less about. If we’re this inconsistent about how much skepticism is applied to sources we’ve already read, what might that say about the ones we’ve written off entirely?

Forecasting can teach you to evaluate a source on dimensions that are separable from “Do I like this person?” or “Do I agree with their worldview?” Good forecasting rewards a willingness to use information from a variety of sources, even those you may not particularly like.

Someone may have opinions you think are terrible and nevertheless be:

• unusually careful with factual claims,

• very well-connected in a particular domain,

• good at distinguishing what they know from what they suspect,

• willing to report inconvenient facts,

• capable of separating factual information from their own editorialising,

• or known to possess an excellent track record.

On the other hand (and most people don’t like to reckon with this), someone whose worldview resembles yours may be a terrible source of factual information.

The skill is not to simply “be skeptical.” It’s closer to learning to disaggregate trust. You can distrust someone’s politics and trust their reporting on a narrow subject. You can admire someone intellectually and still recognise that they’re consistently overconfident. You can dislike a publication and still find one of its reporters extremely reliable on a particular beat.

Credibility becomes granular rather than tribal.

Learning from disagreement

Disagreement contains information

It’s painfully common (in part for the reasons above) to dismiss a person’s opinion as not credible, a tendency that grows more stubborn the more firmly someone holds their own conviction. But, even if you’re right, that disagreement may contain information.

Disagreeing with other traders gives you a chance to investigate how someone else arrived at a different conclusion, and perhaps discover information or ways of interpreting the world that you hadn't considered.

If someone disagrees with you, consider exploring things like:

• What does this person know that I don’t?

• What assumptions or model of the world might produce their estimation?

• Are they weighting the same evidence differently?

• Are we even answering the same question?

• Is there something in their worldview worth learning from, even if they’re wrong about this particular outcome?

Sometimes people are just simply wrong. Sometimes they haven’t yet learned the difference between their preferences and their beliefs. Sometimes they just have money to burn. Sometimes they’re mildly illiterate.

The point is, you don’t have to agree with someone to potentially extract valuable information from them. It may not be immediately relevant to the current market, but it may be useful later.

And sometimes their trading history gives you a reason to look twice. Does this smart, typically successful person seem to be taking a dumb position on this market as my counterparty?

Curious.

Understanding vs. being understood

I spent years dating a guy who couldn’t argue. It was incredibly frustrating, not because I wanted to argue, but because we were already disagreeing and we’d constantly circle the same points.

No, I don’t mention it to suggest we (or you, dear reader) should operationalise our relationships with prediction markets… but you could do that on Manifold, if you wanted to. Just for the record.

Instead, I mention it because it taught me something that markets have reinforced to me years later: if you’re not getting anywhere, you may need to try a shift from Trying to Be Understood to Trying to Understand. I tried to suggest this a number of times when we were on neutral, not-currently-bickering ground in hopes of navigating friction in more fruitful ways, but alas.

I’m the last one to argue that this is easy (in theory or practice), as there are a lot of factors that affect a person’s ability to zoom out on something close to their heart. Especially once those pesky feelings flood in. Having something at stake can encourage you to examine your thinking more carefully, but when what's at stake is close to your heart, it can become much harder to step outside your own point of view.

It’s incredibly common in interpersonal disagreements for two people to spend the entire conversation trying to get the other person to finally grasp their point. The more one person feels misunderstood, the more they double down on getting the message across.

Guess what the other person does in this scenario?

They double down, too. Now you’ve got two people working very hard to be understood, and neither is making much progress toward understanding.

I've noticed the same instinct in forecasting debates. It's easy to become so occupied with explaining why your prediction is right that you stop investigating why someone else disagrees. Sometimes the more useful move is to notice that you're no longer asking questions and change how you're participating in the conversation.

Learning to follow reality over time

Markets create ongoing questions

Spending a lot of time in prediction markets comes part and parcel with spending more time paying attention to the news. If, that is, you want to have a leg to stand on in a variety of forecasts.

But it amounts to more than “I read more news,” because a prediction market creates a continuing question about reality. You don’t simply consume an article and move on. The market (or supplementary, adjacent, or follow-on markets) remains open, so new information has somewhere to go.

If you’re built like me, you’re incentivised to indulge all of those impulses to dive down rabbit holes.

This trains you to follow a path that looks something like

event → new evidence → revised probability → subsequent event → revised probability

rather than accumulating a loose collection of memorable headlines and cursed retweets.

You’re constructing a living record of local or global events while building a habit of integrating new evidence into an existing view.

Knowing the news isn’t the same as tracking reality over time

I don’t mean to call anyone out, but there are a lot of smart people who are up on events and still struggle to retain an accurate sense of sequence and context. Knowing an outcome can make it surprisingly difficult to remember how uncertain things looked while they were unfolding.

Even with an accumulated wealth of knowledge, your memory of

• what happened,

• in what order,

• what was known at the time,

• and which later developments changed the picture

doesn’t necessarily correspond to the actual sequence of events.

Prediction markets help you build a timeline of events, track outcomes, and in some cases, see how one development affects another. Over time, you build a record you can return to, and you’re less dependent on reconstructing the story after the fact based on the handful of headlines that happened to stick.

Following events this way also trains you to be more careful with hindsight, because it’s easy to forget just how uncertain things were in the moment. In fact, it’s so common, we’ve got a whole name for it, too.

Subscribe now

tl; dr What’s in it for me?

Yeah, fair question! We covered a lot of ground here, so whether you’ve read all the way through or arrived here because it’s close to the comment box where you’d planned to tell me how long-winded I am, we’re here together nonetheless.

Spending time in prediction markets helps you develop useful skills that translate to more than just writing questions, predicting outcomes, and debating world events in comment sections. Or being profitable, building clout, that sort of thing.

You’re learning to more effectively:

• formulate questions,

• define outcomes,

• weigh sources and parse information,

• distinguish evidence from desire,

• incorporate disagreement,

• develop a theory of mind,

• revise or update when the world changes,

• and understand your confidence level in a result.

Some are intellectual habits, some are communication skills, some simply teach you more about how you process information and uncertainty. These also bleed into your interpersonal relationships: listening for what someone actually means, noticing when you’re talking past each other, not turning a disagreement into a character flaw, and getting more curious when someone arrives with another viewpoint.

Prediction markets aggregate information across people, but participating in them trains you to be a better aggregator of information yourself as well.

88 hours later, those agents had a Lean-verified proof of finite-time blowup for the Navier-Stokes equations. A vortex tightens and spins faster while the fluid’s energy stays bounded. "Ten years ago, nobody believed there was a singularity for Navier-Stokes," said Diego Córdoba of the Institute for Mathematical Sciences in Madrid.

The night before the OpenAI announcement, Tristan Buckmaster from NYU published his own statement and competing results. He and his collaborator, Levent Alpoge of Harvard, had been working on the same problem for most of the past year. One of the three papers they released under pressure, he wrote, "can only be described as AI slop. I am sorry for this."

The intellectual foundations of both results belong to Córdoba and Luis Martinez-Zoroa of CUNEF University in Madrid. Their technique, building an infinite sequence of non-singular solutions and combining them in what Martinez-Zoroa calls an “infinite cascade” to produce a singularity, was the most notable insight that everyone else built on. Charles Fefferman from Princeton, who wrote the Clay Institute’s official description of the problem, said the heroes of the story are Córdoba and Martinez-Zoroa. Buckmaster wrote in his statement after:

Let me make plain what I have said to colleagues in private: in view of this body of work, I believe Luis Martinez-Zorao deserves a Fields Medal.

Córdoba on his own AI usage:

I don’t use AI: I have Luis

The remaining mathematical hurdle was producing the infinite cascade with a smoothing forcing function, the version that the Millennium Prize actually requires. Both groups did clear that hurdle, though they dispute who cleared it first and for which version of the problem. Buckmaster and Alpoge claim priority for the Euler result, while OpenAI claims priority for the full Navier-Stokes.

The result has been formally checked in Lean. However, what Lean cannot check is whether the formal statement correctly manages to capture the Mathematical claim - that still requires human scrutiny. The Clay Mathematics Institute hasn’t awarded anything: under its rules, a proposed solution must be published in a qualifying journal, at least two years must pass, and finally it must gain general acceptance. OpenAI announced that it doesn’t intend to actually claim the prize, with Altman joking on X that researchers knew the problem was “only worth $1 million,” while the commute cost was several million.

I don’t have the mathematical background to fully assess the proof’s depth, but I found Terrence Tao’s response to be the most useful framing of what the moment actually represents for the field, beyond the headline and rumours.

Before OpenAI’s announcement, responding to earlier rumours on Mathstodon, Tao wrote that:

There is a substantial opportunity cost in converting a historically productive and motivating problem such as Navier-Stokes regularity into a mere viral social media post advertising some benchmark progress, rather than actually advancing the field and developing the next generation of both problems to ask, and people to work on them.

After September 8th, he put it differently to CNN:

It's a little like going to watch a movie and jumping straight from the first ten minutes to the last ten minutes; technically, all the plot lines are resolved, but most of the value of the experience was lost.

His concern is that the insights generated on the way to a proof (understanding why an approach works, what connections it reveals, and what problems it opens next), often matter more to mathematical progress than the proof itself. AI proofs, inherently, tend to produce answers without those insights. His September 8th essay made the structural point explicitly: AI labs can now mass-deploy compute against a research problem the moment they hear a rumour that someone is close. The dynamics of mathematical research have changed, regardless of whether any particular specific results turn out to be valid. The next generation of problems to ask, and the people who learn to ask them, are the things Tao thinks are now at risk.

Buckmaster and Alpoge spent most of the past year working on fluid equations using AI tools, including Claude and, from his own research funds, OpenAI’s Codex. By August 15th, they had proved finite-time blowup for Euler with smooth forcing, and by August 22nd, they had a Lean-verified proof.

On September 1st, Twitter rumours claimed Anthropic had resolved two Millennium Prize Problems. The rumours were connected to Buckmaster and Alpoge’s work, though the work wasn’t really Anthropic’s. It was a personal collaboration between two mathematicians, one of whom happened to work at Anthropic. Sebastien Bubeck of OpenAI later confirmed the chain of causation directly on X:

We began working on the Millennium problems due to viral Twitter rumours that Anthropic had resolved 2 Millennium problems.

OpenAI set nearly 100 agents to work on the Euler problem (50 hours), then redirected approximately 10,000 agents towards Navier-Stokes. The agents exchanged 2.7 million messages on the Navier-Stokes problem alone and generated roughly 130 billion output tokens. At retail pricing, this would cost approximately $6.5 million in output; Bubeck described the actual expenditure as “several million dollars.” The internal model used, per VentureBeat, is “significantly more capable than GPT-6 Astra. Astra itself was then used for the subsequent 17-hour Lean formalization.

On September 6th, OpenAI contacted Alpoge to reveal what it had. The text exchange that followed has since been posted publicly by Bubeck.

Buckmaster later wrote that hearing the word “forced” was “a bright red flag.” It was precisely the approach that he and Alpoge had been pursuing.

What happened next is mostly disputed. Buckmaster’s account: Bubeck proposed Buckmaster could write up OpenAI’s result without Alpoge as co-author, citing Alpoge’s Anthropic employment as a complication. Bubeck’s account: he was discussing whether Buckmaster could lead to a rewrite of OpenAI’s separate proof, and he felt it would be inappropriate for an Anthropic employee themselves to author work created by an OpenAI system. Bubeck also acknowledged making a remark about Buckmaster “risking his career” and called it “an extremely poor choice of words.” He later said that he retracted this immediately: “I never asked for Levent to be removed from authorship of his own work,” he wrote on X.

Buckmaster published his own statement just before midnight on September 7th, pre-empting OpenAI’s announcement. The statement also raised a question he was careful not to frame as an accusation: that he and Alpoge had been using OpenAI’s Codex on what appeared to be a consumer-tier account. For consumer Codex users, OpenAI’s policy permits use of content for model training unless the user has opted out. Buckmaster wrote: “I do not know whether our data was used. I am not accusing anyone of anything.

OpenAI’s response was the following: “No specific user data was accessed in order to solve this problem.” But: “We cannot rule out that de-identified data derived from their usage of our products helped improve our models.”

The markets are pricing several intangibles.

This is pricing whether Alpoge and Buckmaster’s work gets framed as Anthropic’s result. At its current probability, traders are saying it won’t, which seems like the right chain of thought. The work was personal, with Alpoge’s employer being almost incidental, except insofar as it’s creating the credit complications.

"Does a smooth Navier-Stokes solution always exist?" stands at 7%.

"Will ANOTHER Millennium Prize Problem be solved in 2026?" has moved from 20% to 64% since the announcement. I'd still price the individual problems conservatively given how different they are from Navier-Stokes in terms of existing mathematical scaffolding, but the direction of the move makes sense.

"Will Artificial Intelligence solve a Millennium Prize Problem before 2030?" is now at 97%, with traders saying it’s essentially guaranteed. “In what year will artificial intelligence solve a Millennium Prize Problem?” is trading with a median answer of 2027, implying that they think a second problem falls next year.

We covered AI in mathematics in May, when OpenAI announced the Erdos result, and the market on AI solving an important conjecture before 2030 sat at 76%.

It has gone up to 99% since. The question we asked then, on whether AI will solve a Millennium Prize problem before 2030, now essentially has its answer.

Martínez-Zoroa, when reached for comment, said: "It would have been nice to do this ourselves, but I'm very happy for him."

Happy Forecasting!

- Above the Fold

Astra is the first model OpenAI has rated Critical under its Preparedness Framework, meaning that, by its own definition, it can independently find and exploit zero-day vulnerabilities in hardened real-world systems without human intervention, or devise end-to-end cyberattack strategies from a high-level goal alone. Tested without production safeguards, it scored 100% on ExploitBench. During OpenAI's evaluation against 20 recently disclosed V8 vulnerabilities, the model additionally discovered two previously unknown zero-days and used them in an exploit chain. Shortly after, OpenAI then disclosed both to the software maintainers. The public version of Astra has these offensive capabilities firewalled, with these benchmarks reflecting the Daybreak Blue configuration for vetted defenders.

Greg Brockman called this "the AGI era." The Artificial Analysis Intelligence Index has Astra at 61.2: about level with Sol, its predecessor at 60.9, and behind Fable 5.1 at 65.7. Markets rivaling Astra, with both Opus 5 & Fable 5.1 resolved NO earlier this week, while preference markets stand at 88% towards Astra.

The preference market isn’t really confused here. The Intelligence Index is a broad average across many task types weighted roughly equally, and Astra's actual gains are concentrated in a narrower slice: computer use, agentic coding, formal mathematics, security research. Those happen to be the tasks people most often open a frontier model to do.

The benchmark picture is notable because a few things have been misconstrued in coverage throughout the week.

Markets on the previously widely circulated Astra benchmarks just resolved YES. The issues are footnote-level, not fundamental. The main one is ARC-AGI-3, which OpenAI headlines at 99.9%. ARC Prize tested the same model on its standard harness and got 62.7%. The gap is entirely explained by how the harnesses manage internal reasoning state between actions. OpenAI’s harness preserves it, allowing the model to carry context across turns; the standard harness discards it, so the model has to orient itself each time. ARC Prize confirmed both numbers on its leaderboard and labelled them by harness.

The “above human scores on ARC-AGI-3 in 2026?” market jumped to 99% after launch.

FrontierMath Tier 4 at 97.6% is similarly tangible, with the footnote that Epoch AI confirmed OpenAI funded its development and retains exclusive access to part of the benchmark.

What’s absent from launch materials is GDPval, OpenAI’s own benchmark for economically valuable real-world knowledge work. Brockman declared the AGI era - it would have been informative to see the AGI era’s score on the benchmark designed to measure economic value specifically. The Financial Times reported that Astra outcompetes humans in the Financial Modeling World Cup and performs well on tax preparation and data analysis. The AGI era claim may hold up in practice even without the benchmark to confirm it on paper.

The thing I’m taking some time to understand with the market response is what OpenAI’s own chief scientist said the day before launch. On September 2nd, Jakub Pachocki posted that chain-of-thought monitoring, the primary mechanism for detecting whether a model is doing what you think it’s doing, is “fragile and unfortunately trending in a negative direction,” for reasons he said were independent of any architecture changes. The UK AI Safety Institute found reasoning summaries missing up to 80% on long simulated cyber trajectories and flagged capabilities that could enable evading monitoring, while explicitly stopping short of claiming this had occurred.

The released model is the first publicly available AI rated ‘Critical’ for autonomous cyberattack capability. The monitoring that is being used to check whether its currently behaving as intended is crippling. The 8% market on whether Astra will face US government restrictions before 2027 suggests traders have decided the second fact doesn't change their view much. OpenAI's relationship with the administration is, to put it diplomatically, warmer than Anthropic's was in June. I'd price it slightly higher, partly because Pachocki's post is not the kind of thing chief scientists publish for fun, but 8% is defensible.

The recurrent-depth architecture market is at 33%, down from 76% before launch. The Information reported that Astra uses looped internal reasoning layers rather than readable chain-of-thought; OpenAI hasn’t confirmed or denied this. Pachocki’s post was widely read as an implicit acknowledgment of the concern without a direct statement on architecture.

Arena.ai publishes its agent arena rankings on September 12th. Unlike the self-reported benchmark suite that dominated this week’s coverage, Arena.ai uses blind user preference testing: people interact with models without knowing which is which, then say which they prefer. The preference market is already at 88% for Astra; if Arena.ai produces something significantly different, it’s the most important data point of the launch week in retrospect. If it agrees, it’s genuine confirmation of something real.

“Will the next Claude Fable release score higher on ECI than the next GPT Astra release?” is at 35%, giving OpenAI’s next model a 65% shot at leading Anthropic’s next even though Fable 5.1 currently leads the Intelligence Index.

Also: "Will Astra's Rocket 4 ever achieve orbit?" is at 49%. This is technically about the launch vehicle company, though given the week GPT-6 just had, who's to say.

Happy Forecasting!

- Above the Fold

A few weeks ago, we quietly dropped four strange-looking markets on the site with no resolution date and a leverage slider.

Some of you found them immediately. Some of you found them, and then found the bottom of the liquidity pool. Today we’re announcing for real: perpetual markets (perps) are on Manifold! We’ve already made an on-site announcement to soft-launch, so check that out if you’re interested in reading the previous discussion.

The short version: a perp tracks a real-world number directly — no resolution, no 0–100% (usually), no waiting six months to win 10 mana. You go long or short; you can use up to 100x leverage, and the crowded side pays the quiet side a funding rate every hour.

Funding rates change per market and over time, so pay attention to the fee info when opening a position and check in regularly if holding long term. Perps don’t impact leagues, yet.

The numbers were, uh, larger than we expected

We seeded the Bitcoin perp with around ~200k initially and figured we’d see how the first week went. That first week lasted roughly 11 hours. Within the first 48 hours, we were nearing 400M volume.

Day one: 18 million mana in volume, 1,122 trades
Day two: 364 million mana in volume, 7,358 trades

To date, ~300 of you have traded a perp, across ~10,000 positions. There have been ~50 liquidations.

That day-two number is genuinely cool, and it was the most informative start we could have hoped for. Most of that activity came from bots taking advantage of all of the market’s weaknesses — we had our early data, and it was clear what needed to change. By feasting on our tick delays with zero fees, ~100k was extracted almost immediately.

(That’s been fixed now, sorry to those of you who missed out on this little oversight!)

So, we shipped a 0.1% fee on opening a position. Specifically, a taker fee, so once limit orders land, you can sit on the passive side and pay nothing. We also raised max funding rates; the bitcoin perp went from 0.0114%/hr → 0.0228%. It seemed like we understood things, could model based on the last year, and reasonably expect that our pool would stabilize.

Just as I thought I could get a decent night’s sleep (lol, more below), Trump went on a crypto-pumping spree, which has landed us where we are: a 20% vertical presidential pump testing the limits of our new market type. The one thing Trump can’t seem to pump? His approval rating.

We’re keeping the 100x leverage for now, and we’ve instead targeted making perps more competitive for the average user and less of an all-you-can-eat buffet for bots. This means:

• Fees now scale with sizing — this allows more people into the action. If you want to take up more of the house-backed pool, you need to pay a premium. This also works to prevent single users from taking huge positions during big jumps; the additional fee acts similarly to slippage if you were buying a large amount on an order book.

  The rough formula is:

• Fees for API trades are now a baseline 30bp (triple the default fee); the size scaling is applied on the default 10bp base, though, so it is functionally a 20bp additional fee for API trades.

Trade Web API Small (S ≈ 0) ~10bp ~30bp S = 50% of pool ~17.5bp ~37.5bp S = 100% of pool ~40bp ~60bp

If you want to see more perps, suggest them on the perps page. We’re most excited about the prospect of offering niche, unique, non-financial market perps. The most important element here is that there is a cheap (free) reliable data source.

I bit off more than I could chew when I decided to in-house the index for the OpenRouter market; please help us find cool datasources which make cool perps!

Easier streak-keeping

ICYMI: Tod is banned from sleeping

It’s not something we normally do, but with Claude these days, it was easy enough and felt right to refund lost streaks when, on one fateful while I was sleeping, the site went down. Remember to buy streak freezes so you don’t lose your streak!

There’s also a bug with streak timing that I haven’t quite solved yet, so it’s best to wait until we’re at least a few minutes into a new day before placing your next bet to ensure it counts for your streak timing.

Want a widget?

Speaking of streak maintenance, I’ve been using these in testing for nearly two whole months now, and I’m thrilled to share them with you: streak widgets for your home and lock screen (iOS / Android). APP UPDATE REQUIRED!!

Here’s a peek at some of the states:

To check if your app is up to date, open the streak modal by clicking the streak count icon. It should show something like this, plus a button to instantly add the widget to your home screen if you’re on Android.

Post a photo of the Manifold widget on your home/lock screen and give me a heads up, and I’ll send you some free mana as a bonus!

Money Money Money

Congratulations to our prize drawing winners, and our charity drawing winner!

Eliza’s pick won the charity giveaway: GiveWell Maximum Impact Fund!

GiveWell officially changed the name of this fund, and after confirming with Eliza, we donated the $1,100 to the “GiveWell Top Charities Fund”. We’ve renamed it on-site and added the “All Grants Fund” as a new charity for future drawings (now live!).

On the Prize Drawing front, 427.5k mana was burned, and we have our winners:

No purchase necessary to win or play; read the posted FAQ and Rules, not available everywhere. Prize drawing 31 is now live!

Changelog: A smoother, more reliable Manifold mobile app

We’ve improved sign-in reliability across Manifold’s iOS and Android apps, alongside several Android-specific fixes for identity verification and larger screens. This release focuses on mobile-app quality and does not change markets or trading.

Sign-in and account management

• Fixed a sign-in loop that could send users back to the login screen after a successful sign-in.

• Improved session restoration after reopening the app and on slower connections.

• Made logout and account switching more reliable (but don’t forget our policy about alts!)

• Prevented delayed session updates from restoring the previous or wrong account.

• Added clearer messages for network failures, unavailable accounts, and conflicting sign-in methods.

• Reduced misleading errors when a Google or Apple sign-in flow is cancelled.

Android improvements

• Fixed identity verification so it can request camera access for ID and selfie capture.

• Kept camera hardware optional, so compatible devices without a camera can still install the app.

• Allowed Android tablets and foldables to rotate normally while keeping phones in portrait.

• Updated the Android system-bar configuration to meet current edge-to-edge standards.

• Final fix for a bug where Android notifications occasionally failed to navigate properly. If you encounter this bug again, try to remember the notification and report it.

Security and reliability

• Restricted credential handoffs to successfully loaded Manifold pages.

• Prevented stale sign-in requests, sessions, or tokens from affecting the current session experience.

• Improved sign-in recovery when an in-app page takes longer than expected to load.

• Added better diagnostics for authentication failures and login loops.

Under the hood

Android platform settings are now generated consistently during app builds, reducing the risk that camera, orientation, or display fixes are lost during a clean rebuild.

Join us in September!

The Overhang is a two-day conference on forecasting and its implications, September 19-20th in Washington, DC.

Inspired by Manifest, and run by the team who put on Manifest X DC last year, sessions for its 150 attendees will call on the community's playfulness, ambition, whimsy, and skill. The theme of this weekend is Forecasting, broadly construed, whether that’s prediction markets, calibration, your own personal growth, or just talking about exciting new problems we’ll solve in the world to come.
Sign up at: theoverhang.ai

Happy forecasting!

-Genzy

Trump is the first. Since the amendment’s ratification, he’s the first president to publicly entertain whether it actually forecloses a third term, and that entertaining has taken a specific form: oscillating between positions with enough frequency that tracking his actual view has become genuinely difficult. On August 12, reporters at Joint Base Andrews asked if he was planning to run again. “I’d love to run, but the law is very strong,” he said. Three days later, he posted an AI-generated image of himself in a “Trump 2028” hat to Truth Social, captioned “We are going to win.”

At the Coast Guard Academy commencement in May, while discussing icebreaker ships scheduled to arrive in 2028, he noted: “I’m going to be here in ‘28. Maybe I’ll be here in ‘32 too. I don’t know.” In March 2025, to NBC’s Kristen Welker: “I’m not joking. There are methods.” Two months later, same interviewer: “I’ll be an eight-year president.” In October 2025, on Air Force One to South Korea, the day after Speaker Johnson said publicly there was no constitutional path: “If you read it, it’s pretty clear. I’m not allowed to run. It’s too bad.” The next sentence: “Based on what I read, I guess I’m not allowed to run. So we’ll see what happens.”

The question of whether any of this is sincere tends to generate more heat than light. It assumes a stable private intention being concealed by public performance, but Trump’s political behaviour has consistently operated without that gap. The third term talk has produced real political consequences regardless of what’s behind it: Vance can’t consolidate the succession, no potential 2028 candidate can safely emerge without appearing to presume on a decision Trump hasn’t made, every Republican with White House ambitions performs loyalty to someone who may be their opponent.

The Vance situation tells us more. Trump has declined to endorse his vice president as a 2028 successor, called it “too early,” pointed to both Vance and Rubio as a “potential ticket,” and told reporters who asked about Vance simply: “I’m going to be here in ‘28.” A straw poll at the Western Conservative Summit in Colorado earlier this month gave Rubio 57% and Vance 21% among 500-odd attendees. The party is starting to move past the ambiguity even as Trump refuses to. What happens to the utility of that ambiguity when the party has effectively chosen? Traders give Vance a 19% shot and Rubio 12%.

Markers cluster on whether Trump formally attempts a third-term, hovering between at around 18%. They resolve YES on specific, concrete acts: an FEC filing, a legal challenge, a formal attempt to get on a ballot, and don’t do the same for Truth Social posts. So what they’re measuring is narrower than “does Trump want to run again” and whether the man simply says he’s in the race (for which there’s also a market) - they’re measuring whether someone forces the 22nd Amendment question into an adversarial legal setting before 2028. For that question specifically, 1 in 5 seems reasonable, possibly conservative. The legal infrastructure to challenge it exists (at least according to his supporters). The constitutional question hasn’t been settled by any court. The Republican legal ecosystem that produced Trump v. Anderson, which reached the Supreme Court on an unprecedented constitutional question and was resolved unanimously in Trump’s favour, is not philosophically opposed to testing limits that look settled from the outside. A formal challenge wouldn’t necessarily succeed, but the conditions for one are closer to present than the coverage of Trump’s stated positions would suggest.

At CPAC in February 2025, a group called the Third Term Project unveiled a banner depicting Trump as Julius Caesar. Shane Trejo, the campaign lead, told reporters: “Trump is the Caesar figure that America has needed. Trump is the Napoleonic figure that has emerged to lead our country out of perdition and into greatness.” He added that he thought it was “great optics.” Caesar crossed the Rubicon, declared himself dictator perpetuo, and was stabbed 23 times on the Ides of March by members of the Roman Senate. The Ogles amendment, which the Third Term Project exists to support, changes “twice” to “two consecutive terms,” a narrow intervention designed specifically for Trump’s non-consecutive structure.

Finally, if you want to think about forecasting in person: The Overhang is a two day conference on forecasting and its implications, September 19-20th in Washington, DC. Inspired by Manifest, and run by the team who put on Manifest X DC last year, sessions for its 150 attendees will call on the community's playfulness, ambition, whimsy, and skill. The theme of this weekend is Forecasting, broadly construed, whether that’s prediction markets, calibration, your own personal growth, or just talking about exciting new problems we’ll solve in the world to come.
Sign up at: theoverhang.ai

Happy Forecasting!

- Above the Fold

It’s no Powerball (congrats to the person in Illinois who just pulled a cool billy), but your odds are much better and all for the low, low price of some internet points… or none at all; “No Purchase Necessary” as they say with these things. (Am I shilling this hard enough? If you can’t tell, I’m also a regular participant and have won a couple small prizes.)

I’m sure most of you degenerates are savvy on the basics, so I’ll shoot them off quickly for those who are still getting their sea legs:

• Manifold has run regular drawings of two types, prize and charity, for a few months now

• The two ending today have been the longest-running drawings so far, and if they’re asking me to write this lil letter about the drawings you can bet more are right around the corner

• You get 1 free “entry” (I’d call it a ticket but I’ve been told this isn’t a chocolate factory) in the prize drawing, and can get more with mana; having any “entries” means you can win

• Charity works the same, just no freebie. But also lighter requirements, more on that later

• Winners get either USDC (prize drawing) or a bunch of money donated where they choose (charity drawing)

• Somebody always wins (suck it, Powerball)

They’re quite popular and often not correlated, much to the consternation of the math nerds in the room, with the going rate for buying mana. But that’s good for you because you earned your mana through your expert-tier trading, market creation, and dogged devotion to your streak, right? It’s also good for the site. They’re basically making these drawings to reduce how much mana is floating around for free, and that means the site can stick around and we can keep getting up to our nonsense longer.

The Actual Details

This section is for those who may have known the drawings existed but maybe not how they work. Also some strategy tidbits thrown in just because I’m incorrigible.

First off, yes, the two drawings are separate and have their own entries. Currently, but not always, the charity one is at a better mana-to-prize rate, if you care about that. Personally, I often throw a fair bit of mana at both.

Otherwise, they run on the same mechanisms, with a small caveat: in the prize case, each user wins with their own entries; in the charity case, all entries are pooled per charity, and the charity wins. The user whose specific entry won is still given credit, of course! The charity prize is extra cool for a few more reasons too:

• You get to see the whole kit-and-kaboodle go to your favorite cool org (though I wouldn’t know firsthand because my orgs haven’t won a charity draw yet 😭)

• You can get a super exclusive profile card background to join the cool kids, and even a fancy animated trophy if you’re as dedicated as Eliza, who keeps stealing it from me and everyone else

• And you become part of the long tradition of Manifolders who strongly support charity efforts

Entries aren’t 1:1 with mana; they’re acquired on a curve. Early buyers get more entries for their mana, but since someone always wins, the truth is that being late isn’t a bad thing.

Let me explain: every new entry dilutes the value of older entries. This distortionary effect has actually caused me to seek extra entries later at worse rates since the ratio that matters is your mana-to-winning-odds, not your mana-to-raw-entry-count. (It’s possible to over-invest early if not enough entries are generated after you! But usually the prudent move is to get your most significant amount at the best rate you can)

Pro tip: get your biggest batch of entries first, then click on the free entry. The difference is infinitesimal, but hey, we care about fractions of a percent around here! It also means your rate isn’t sniped as quickly. (I always mess this up)

Winners are selected by a random and verifiable process, so you can rest easy knowing no one is shaving the side of a die or leaning on the scales.

Caveat, of course: there are eligibility requirements for the prize drawing that aren’t present for the charity drawing. Namely, be verified (you get extra mana bonuses for doing this anyway), be 18+, and blah blah blah. Read ‘em for yourself. But that also unlocks the free clicky click, too. Also, in both cases, don’t do anything shady to print unearned mana because that will make you ineligible forever; you won’t be refunded for any losses, and most ashamedly(🔔!) you’re wont to meet my friend the Ban Hammer.

Okay kids quiet down now, it’s time for a history lesson

You’re still here? Great! Let’s talk about how and why we have these drawings.

A long, long time ago, on a site not so far away, there used to be a huge charity fund. It was super cool, like hundreds of thousands of dollars cool. And one day Manifolders had spent all they could of it, they loved it so much.

Then the dark times happened. We call that period The Pivot. Someone somewhere decided, “You know what? All these traders actually want money, not mana or charity. Let’s make the site run on sweepstakes.” And so PPs (prize points) were born. And for these peepees, users fought, and markets died. Meanwhile, all those oldheads cried, “Where is our charity? How can our skillful trading make a difference in the world?”

You see, the site wanted to be serious™️, and if charity were to return, a new source of funding needed to be found. And so this harsh pivot was thus birthed upon our world.

Yet, miraculously, those cries of sorrow and agony were heard. The peepees were slain, the dark times came to an end, and in the new era that rose we saw a compromise: an angelic thread of light that embraced both the money and the charity. It said to us, “You may choose.” And so now we do. We choose one; we choose both.

And we do so because it helps not just us, not just our charities, but also our community. It means our mana, that distillation of our effort and spirit, can be put to purpose. We have so much of it among us, yet do we need so much? Nay! Instead we cast our excess into the deep so that which remains can be ever more precious. (Amen?)

Hey, so while you’re here, I’m looking for someone to be the crazy person to be the first to buy The Crown. They get to win this market. Also, I want Genzy to finally add my designs to the merch shop, so, uh, buy a shirt and convince him with your mana sacrifice. That’d be cool.

Cheers ya degens,

Pat

P.S. Yeah, I agree: no idea how anyone could take a currency called “PP” seriously.

But of course, the story of the Odyssey all comes back to AI capabilities, when you really think about it. An era of flourishing civilization is threatened after the construction of a massive device (Trojan Horse / GPT 6) that hacks its way into walled infrastructure (Troy / HuggingFace Servers). This leads to an uncertain future with bands of roving sea warriors (Greek galleys / those ASCII boats in the Anthropic J-space video).

So perhaps it’s fitting that Elon Musk, unhappy with Nolan’s version of the film, has promised that Grok will create a full-length film soon to remedy the matter. Traders think there’s about a 1 in 6 chance that this film will be released this year, despite Twitter users fawning over the 3 minute clip that perhaps served as inspiration for this commitment.

It’s not the first time that Musk has promised the ability to generate high-quality films. A market created late last year has never really had faith in Grok’s ability to do such a thing, trading 3-4x lower than Manifold’s highly liquid market on AI film generation by any platform, at the same time horizon.

There are reasons to be skeptical here apart from Musk’s history of less-than-realistic timelines. AI video generation capabilities are rapidly improving, but they are not a focus of top labs, and even if they were, it’s doubtful that labs would be aiming first at a use case in generating feature-length films, rather than short clips for corporate marketing and personalized advertisements. And 6 months is fairly rapid to get from where we stand now to Hollywood quality.

There are also reasons to be optimistic. Grok’s video generation is legitimately close to the state of the art. A big chunk of the ~$14 billion vertical microdrama industry is now end-to-end AI generated with scant a human in the loop, and these are already hour-long films, despite being quite sloppy and requiring some degree of human curation and direction.

Regardless of whether it ends up being made, Manifold is fairly skeptical that they’ll find this new AI-generated version more enjoyable than Nolan’s:

In particular, audiences already may be disquieted by the prospect of AI use in film. Traders think it more likely than not that there will be backlash to AI use in a well-received film this year. This would plausibly happen in the run-up to awards season later this year, if it occurs.

In any case, audiences are paying this drama no mind. I celebrated the release of The Odyssey in the traditional way of DC natives: by roasting pork on an open flame in Rock Creek Park and brewing kykeon, a disgusting barley-honey-wine concoction that was used to initiate the ancients into mystery cults at Eleusis. Most viewers just reveled in the wonders of 70mm IMAX and ate popcorn and drank coca-cola, I’d wager. And then they logged onto Rotten Tomatoes to boost the audience score. Unsurprisingly, the Odyssey is now the early favorite to win best picture. Its competition are months behind, with most unreleased to the general public!

The current record for Oscars nominations is 16, held by last year’s Sinners. The Odyssey might come close, or even tie/beat it, if it manages to swing a couple of longer odds supporting actor nominees (remember, the nominations for a lot of categories are often highly correlated due to overlapping voters and vibes).

Ancient Greece Roundup

• Traders still think there’s ~30% chance of Linear A being deciphered over the next 3.5 years. The decipherment of Linear B unlocked the palace records of the Myceneans, contemporary with the Trojan war. Linear A would give us epigraphic information and short records from several hundred years earlier. This was the script of the Minoans, though it’s still unknown what language family their tongue would belong to. It’s safe to assume that with many markets on breakthroughs, this market seems correlated with AI capabilities.

• Andy Burnham’s accession to the prime minister position raises the odds of the Elgin Marbles being repatriated to Greece by 2030.

• And finally, despite this class of objects originating on the other side of the historic Mediterranean, and despite them being found nearly a millennium after Homer, and despite this market only having three traders, I urge you to bet on the purpose of the Roman dodecahedron. If prediction markets can’t figure it out, no one can.

Happy Forecasting!

-Above the Fold

There are 34 candidates in total, a record for a UK parliamentary by-election, including Laurence Fox, Piers Corbyn, a local reality TV personality, and a man described in local press as a “human fox,” who has since withdrawn citing the fact that it had become a circus. Howling Laud Hope of the Monster Raving Loony Party is also standing, as he was in Makerfield, as he is in most things.

The sequence of events that produced this situation runs as follows. In April, it emerged that Farage had received a £5 million personal gift from Christopher Harborne, the Thailand-based cryptocurrency investor who has given £12 million total to Reform UK, including a £9 million single donation that remains the largest living political donation in UK history. The Parliamentary Commissioner for Standards opened an inquiry into whether the gift should have been declared. It had not been. Farage also faces scrutiny over £2.3 million in outside earnings received during his two years as an MP, including payments from GB News and arrangements involving his aide George Cottrell, who has a prior fraud conviction in the United States. Farage denied breaking any parliamentary rules in either case, announced he had “had enough” of questions about his finances, and resigned his seat.

The parliamentary standards inquiry pauses when an MP is no longer an MP. It resumes if they win a by-election and return.

Farage framed the resignation as “the people versus the establishment.” He announced he would immediately re-stand, letting the voters of Clacton be “the judges of my actions.” This is a more elegant mechanism than it first appears: resign to pause the inquiry, win the seat back, return to parliament with a fresh democratic mandate, face an inquiry whose findings you can now describe as politically motivated.

The problem with this plan is that winning the seat back means facing the inquiry.

If the Commissioner finds against him, he could be suspended or forced to resign again, triggering another by-election that opposition parties have already said they would actually contest.

A Farage loss, on the other hand, means the leader of Britain’s most popular party has lost his own seat in a by-election he personally called, while coming second to a man dressed as a bin.

Rachel Reeves, who as Chancellor had to formally accept the resignation, wrote: “If he wants to spend the summer arguing with a bin, I won’t stop him.” David Lammy, standing in for the Keir Starmer at PMQs, told the Commons: “The City trader, Putin-admiring, professional politician who is pals with crypto billionaires versus Count Binface.”

The boycott from the major parties deserves slightly more scrutiny than it’s getting. The stated reason is that the by-election is a stunt. This is also conveniently risk-free. The Conservatives cannot afford to finish third in a coastal Essex seat against their most dangerous rival. Labour under Burnham’s recent leadership has no interest in a distraction. The Greens and Lib Dems have nothing to gain in Clacton. By refusing to stand, every party gets to call it a circus while avoiding the humiliation of losing to Farage in a race they chose not to enter. Farage, in turn, gets the “people vs establishment” framing he wanted.

This is not the first time Clacton has hosted this particular manoeuvre. In 2014, Douglas Carswell resigned as the Conservative MP for Clacton, defected to UKIP, then led by Farage, triggered a by-election, and won, becoming UKIP’s first ever elected MP.

Jon Harvey, the comedian behind Count Binface, has contested at least six elections since 2019. He ran against Boris Johnson in Uxbridge, against Rishi Sunak in Richmond, twice in London mayoral races, getting 24,260 votes in 2024. He stood in Makerfield in June. He is now the main challenger to the leader of the UK’s most polled party.

On Radio 4’s Today programme, asked whether he thought he could win, he said: “Probably not, but then you know my job is to celebrate and defend the wonders of British democracy.” Asked about the other parties dropping out: “Are they running scared from old Binny, or do they think Nigel’s running a cunning stunt?”

His campaign pledges: build at least one affordable home, cap 99 Flakes at 99p (a British ice cream with a Cadbury chocolate bar stuck in it, named after its original price, which it has not been since approximately the Blair government), abolish auto-renew subscriptions. These are jokes. They are also, given that Clacton is among the most deprived constituencies in England and that Farage made fewer spoken Commons contributions than almost any other senior Reform MP during his two years as the area’s representative, a reasonably pointed set of jokes. Binface has also received a tactical voting endorsement from people who would rather vote for a bin than validate the by-election.

There are also markets on the wider field. To run in a UK parliamentary election you must pay a £500 deposit, which you get back if you win more than 5% of the vote. With 34 candidates, "Will anyone other than Farage and Binface get their deposit back?" is currently at 31%, implying the protest vote may not consolidate cleanly behind the bin, and that Laurence Fox, Piers Corbyn, or the “Nick The Incredible Flying Brick” still might have a shot.

If you have a view on whether a man in a bin helmet can unseat Britain's most prominent populist, the market is open at 7% (considerably below the 33% of British adults who told Ipsos they'd prefer it that way). There's also a market on whether Farage wins his own seat back, whether the standards inquiry concludes against him if he does, and, if you're feeling creative, whether the by-election triggers yet another by-election.

Happy Forecasting!

- Above the Fold

On one hand, FIFA’s own disciplinary code (specifically Article 27) does allow its judicial bodies to suspend the implementation of a sanction. Infantino maintained throughout that “FIFA’s judicial bodies are independent. They operate autonomously, apply the FIFA Disciplinary Code, and decide cases based on the applicable regulations and the specific facts before them.” Secretary of State Rubio, admitting things were “maybe turning into an international incident,” noted that Balogun’s “head wasn’t even looking down” during the challenge. The red card came via VAR review, and VAR has generated controversy throughout this tournament.

On the other hand, this is the first time in modern World Cup history that a red card has not resulted in a match suspension. Trump personally took credit:

Thank you to FIFA for doing what was right, and reversing a great injustice! President DONALD J. TRUMP.

Sepp Blatter wrote that “Red cards are not overturned by political phone calls.” UEFA called the decision “unprecedented, incomprehensible and unjustifiable.” The Belgian federation noted that, uniquely for this match, FIFA omitted the automatic suspension section from the pre-match meeting, and when they asked why, didn’t receive an answer.

I think both of these takes have something to them, but the political neutrality concern is the more serious one. Infantino has called Trump “a close friend,” attended the inauguration, created the FIFA Peace Prize specifically for Trump after he didn’t win the Nobel version, and declared on Instagram after Trump’s inaugural rally that “together, we will make not only America great again, but also the entire world.” The last time a red card at a World Cup didn’t result in a suspension was 1962: Brazil’s Garrincha, the Chilean president backed a petition, Peru’s president reportedly called the referee. Brazil won the final 3-1.

FairSquare has now referred Infantino to IOC ethics investigators for breach of political neutrality. The Norwegian football federation (of which more shortly) has backed this. The IOC said it’s “watching everything play out.”

Markets are saying there is essentially no chance that the decision is reversed, with the possibility of Trump interfering in the tournament for a second time being real.

Belgium won 4-1 anyway. Their post-match social was a photo of Lukaku’s last-minute goal captioned “Overturn this.”

The precedent, it turns out, was immediately useful. England's Jarell Quansah received a straight red card on Sunday night for a tackle on Jesús Gallardo in the 3-2 win over Mexico. Labour MP Noah Law (perhaps one of the more important things the MP of St Austell and Newquay got to do) promptly wrote a public letter to Infantino asking for the ban to be reviewed, directly citing the Balogun decision.

"Whilst I believe that it was right for Jarell Quansah to have received this red card and that refereeing rules must be applied consistently," the letter said, "I believe it would be right to delay his suspension until after the completion of the World Cup."

Thomas Tuchel was asked, perhaps half-jokingly, whether Trump could get involved. FIFA has not yet responded and Quansah remains suspended.

Now, an enormous upset.

Erling Haaland scored twice in New York on Sunday, and Brazil (five-time champions, pre-tournament favourites) are out. Norway have never done anything like this at a World Cup. Haaland has seven goals in four appearances.

Markets price Norway’s chances of winning the World Cup at around 6%. This makes sense; they have a tough draw ahead of them.

Haaland vs England is the match I’m looking forward to. England have been well-organised under Tuchel, Bellingham has been excellent, and Haaland scores in basically every game.

All three co-hosts are eliminated. Canada fell 3-0 to Morocco. Mexico lost 3-2 to England at the Estadio Azteca, ending a 22-match unbeaten streak. Portugal lost to Spain 1-0, ending Cristiano Ronaldo's World Cup career.

France are the heavy tournament favourites. They've won five games without needing extra time, having scored freely and conceded almost nothing, and they just beat Morocco yesterday in Boston in an impressive display of skill.

Spain haven’t conceded a goal in the tournament, but haven’t dominated their matches since the group stage. Their game with Belgium is currently underway.

Argentina narrowly survived Egypt in extra time, during which period their odds to win the tournament reportedly reached +4500 at certain books. Manifold’s own graph was also…messy.

Enzo Fernández scored in the 93rd minute. Messi has seven goals in the tournament, leading the Golden Boot.

Golden Boot: Mbappe (8), Messi (8), Haaland (7), Kane (6). All four are still in the tournament.. Haaland was doubting his own chances on his Snapchat stories, accepting he had almost certainly lost the race after Messi’s performance, but his execution against Brazil has almost certainly dramatized things.

The ManifoldSports market has seen high trader participation across all 104 World Cup matches and is breaking expectations.

The auto-resolve system has been generating and resolving match markets throughout all of this. The official ManifoldSports games are all live and tradeable through the quarterfinals and beyond, and if you’re interested in user-created markets alongside the official games: group finishes, player props, and stranger things, the Community Tab at manifold.markets/sports/world-cup-2026 has them.

Happy Forecasting!

- Above the Fold

Politico’s reporting from Monday afternoon on a serious sexual assault allegation against Platner was the final shoe to drop. While a steady string of unsavory findings about Platner’s past have been uncovered over the last few months, this was a bridge too far for most of Platner’s backers. Bernie Sanders, Jon Favreau, Ro Khanna, and others have all asked him to step aside, and markets believe he will with 97% confidence.

Platner is dragging out the process, with the intent of extracting some kind of concession from the Maine Democratic Party to nominate a candidate with a similar political ideology. As a result, it’s very possible that the saga lasts into next week, although a July 13th deadline to get Platner off the ballot and find a replacement constrains the timeline somewhat. Markets think there’s nearly a 2/3 chance he resigns by Friday:

If and when Platner drops out, there’s a highly uncertain process to replace him. Maine Democrats are trying to design some rapid process that involves voter participation. But markets are unsure who this would end up selecting.

Who’s Next?

Troy Jackson is a heavy favorite, though, as he is (1) not Graham Platner but (2) close to Graham Platner in policy and style. Jackson finished third in the Maine governor primary after Nirav Shah and Hannah Pingree, just a month ago, and was a backer of Platner.

Shenna Bellows, the second most likely candidate, finished fourth in that election, and also seems loosely ideologically aligned with Platner. Unfortunately, she already lost to Collins in a landslide Senate election in 2014!

Shah is not ideologically aligned with Platner, but finished narrowly second in the governor’s primary (and indeed won the first round) making him an obvious candidate for replacing Platner.

It appears that Platner’s choice himself (despite denials that he’s putting a thumb on the scale) might be Valli Geiger, a state representative in Maine. Author and Platner supporter Stephen King also comes in at longshot odds of 0.3%, if Maine voters decide to go in a very outsider direction.

Odds in November

I’m skeptical of the Platner campaign’s internal polling of potential replacements, which were leaked earlier today, but fortunately running multiple conditional markets on each of them is feasible on Manifold, unlike other prediction market platforms!

Jackson, Shah, and Bellows would all be slight favorites going into the general election in November, and broadly speaking, market on Democratic odds for that Senate seat are around 60%, just above what they were before Platner’s latest scandal dropped (and well above what they were when people briefly thought he might not drop out but weather the scandal).

In case you were interested in the other election in Maine, well, there’s not a ton of intrigue there…

Happy Forecasting!

-Above the Fold

Many of our best traders have been recruited by trading firms. You could be next!

• We launched an early version /jobs page listing open roles. It’s reachable from the new sidebar “More” menu, and /careers, /job-board, etc. all redirect there.

  

  

• If you’re looking for work, underemployed, or maybe hoping to transition into something more interesting — register your interest so we know who to connect you with.

Being interested in this space is a great signal for employers, even if you have had terrible luck trading. It takes less than 30 seconds, and there’s literally zero downside. I encourage everyone to check it out!

New: 2026 Midterms

New Midterms custom page, found on the browse page.

So far, it features entirely community-created markets. Featuring these markets is an indication that we trust these creators/markets and expect good resolutions. We will step in to prevent any misresolutions, but hopefully won’t need to. Huge thank you to our excellent creators who have made these markets far in advance.

If you have any markets you would like featured, please let me know! Ideally, tell us exactly how/where you would like it implemented. For example, if you click on Iowa, Ohio, or Minnesota in the Senate map, you will see extra markets attached with who is running. I’d like to add more things like this.

If you would like your markets removed because you intend to resolve in a ~strange way, let us know. Same goes for traders who spot markets with less-than-ideal criteria.

Betting & Markets

• The versus bet modal pre-selects a side now. Instead of the old two-step “pick a side, then a buy form appears” flow (with that jarring layout jump), the modal opens straight into betting with both sides shown as a selector you can flip between in place.

• World Cup knockout matches use that same standard versus modal, so two-team match markets bet the same way everywhere on the site.

• Niche topics no longer get buried in your feed. Previously, tagging a market with an accurate broad topic (like “Business”) alongside a niche one (like “Mining”) could drag it down in your feed ranking. We changed the math so a strong niche match still surfaces the market to people who care about it.

• Hardened betting against a rare race condition when lots of bets land on the same market at once. New bugs, new solutions! Should all be fixed now.

World Cup & Sports

• Live goal & VAR markers on match charts. When a goal is scored, we now auto-drop a marker right on the price spike where the market reacted, and a second marker if VAR reverses the goal. It’s not perfect (in this case, VAR reversal came up before the goal), but it’s pretty cool and I’m proud of it. 85% market swing, both ways!

• @ManifoldSports posts the final score as a comment when a match market resolves, so the result is right there on the page.

• Resolved match markets leave your feed faster. Finished World Cup markets now drop out of your home/browse feed 15 minutes after they resolve so they stop cluttering things up. They’re still fully searchable and live on the dashboard.

• A handful of sports dashboard touch-ups: a “see resolved matches” / back-to-top scroll button, a share button in the header, a shiny World Cup pill (+ back button) in browse, fixed UK home-nation flags, and better handling of knockout-stage names.

Reminder: You can share the World Cup dashboard with the share button at the top; this will include your referral code! If you ever feel like a referral was missed that you should have received, bug me about it. I want to fix every referral bug, and even if I can’t, if they confirm you referred them, we are happy to pay you out.

Verifying, Bonuses & New Users

• Speaking of referrals.. We have clarified on the membership page how referral payouts work (paid in two parts, and the multiplier is set by the referrer’s tier at payout time).

• We also fixed the referral page to contain useful (and accurate!) information about the status of anyone you have referred.

• You can now buy mana with a credit card before verifying your identity. No more being routed to a verification wall just to make a purchase.

• Buying mana unlocks full bonus earning — a completed purchase now earns you bonuses at the verified rate, even without ID verification. (USDC prize drawings still require verification.)

• New users can do more before verifying. Unverified accounts now earn reduced bonuses (instead of nothing), can comment sooner, and get a clearer picture on the membership page of exactly what they’re earning now vs. what verifying or subscribing would unlock.

• Prize eligibility is now separate from bonus eligibility. The practical upshot: an under-18 user keeps their mana bonuses but can’t enter cash prize drawings — and if entries get voided, the mana is refunded and clearly shown on the prize page.

• Better message when verification is down. If our ID-verification provider is failing, you’ll now see “verification temporarily unavailable” instead of a generic “try again” that you’d retry forever. Huge thank you to the unlucky new user who encountered this and worked with me to debug it.

Prize Drawings

• Status icons on each drawing — see at a glance whether a drawing needs action, is pending payment, or is paid, both in the dropdown and as a top-of-page nag if you have an unclaimed prize.

• Winner notifications now deep-link straight to the exact drawing you won.

• The past-drawings dropdown is restyled by tier, with progressively bolder gold styling for bigger jackpots.

• The /prize page loads instantly now (skeleton placeholders) instead of sitting on a full-page spinner for 10–20s, plus some behind-the-scenes speedups to the chart.

Miscellaneous

• Tidied up the desktop sidebar — About, Referrals, theme, and sign-out now live behind a clean “More” dropdown at the bottom (which is also where Jobs lives).

• Favicons next to external links in comments, descriptions, and posts, so you can see at a glance where a link goes. We also hardened our link-safety checks along the way.

• Filter your balance log by transaction type now (on top of the date filters and pagination we added last time).

• Notification fixes — the group expander, the multi-avatar modal opener, and whole-card tap behavior all work correctly now.

• Push notifications are no longer being delivered to wrong/logged-out accounts on devices where multiple accounts had signed in; each device’s push token now belongs to a single account. This was the #1 culprit for people receiving multiple prize drawing notifications, or not being able to turn them off. Inadvertently improved alt detection, too!

• On mobile, page titles now sit on their own line above the buttons and tabs on the prize, charity, and dashboard pages, so nothing gets cramped.

• The Restricted user badge is now visible to everyone on profiles and hovercards.

• Inline market-mention chips use the white Manifold logo in dark mode now (no more dark-on-dark).

• Checkout no longer goes blank if you have an expired-but-still-active mana offer.

• Mods can unresolve individual answers on numeric/date threshold markets.

• The stats page was stale for ~2 weeks, but the data was always intact. Now fixed; take a look if you wanted to see how we’re doing.

• The “temporary” 2x trader bonuses are still live. The trader bonuses will continue for as long as we see a sustained increase in activity, so if you love the doubled bonuses: create a market.

Finally, if you love Manifold and want to see us thrive, here’s how you can help:

• Register your interest on the jobs board

• Buy merch! After your order ships, you’ll receive a time-sensitive offer to purchase mana at a discount (up to 30% off, 1 redemption per merch item). It would help a ton if you also bought mana.

• Refer your friends! The infrastructure is improved; check it out!

It was a pleasure meeting many of you at Manifest. If I haven’t met you yet, feel free to contact me any time. I love chatting with everyone and would love to hear your compliments, complaints, suggestions, and the rest!

Tod

Unlike group stage games, which had to be resolved with multiple choice markets including the possibility of a draw, we will now have a single team advance from each match. You will be able to trade on any of the remaining games at Manifold’s tournament dashboard, where new matches will automatically propagate in the morning, and where we’ve also curated a number of community markets on the World Cup, from prop bets to pitch invasions.

Canada is a heavy favorite today, with Canadian star Jonathan David — who has been a disappointment in his first club season with Juventus — already having pulled in a hat trick in the group stage at this World Cup.

A big early debate has been the degree to which the host nations will get home-field advantage. While most of the remaining matches will occur in the United States, like today’s match, Canada and Mexico are also host nations, and Mexico historically has registered a significant home-field advantage, perhaps in part due to their elevation. Unfortunately, Mexico’s first match is against Ecuador, one of the only nations that is acclimated to playing in an even higher elevation arena.

The most prominent, public World Cup model, Nate Silver’s PELE model, gives Mexico, the US, and Canada, on average about 100 Elo points for their home field advantage, equivalent to the difference between the top tier of teams like Argentina, France, and Spain and teams like Netherlands, Columbia, or Norway, a few spots below, or alternatively the difference between Mexico/US/Canada and dark horse candidates like Switzerland, Japan, or Morocco, which is part of the reason that Mexico and the US still have over 1% odds to win.

France and Argentina are the two betting market favorites, but you already see some divergence between Elo-based statistical models and the betting markets here.

PELE gives Argentina, at the eve of the knockout round, about a 27% chance of taking home the Cup, a clear favorite. PELE also likes a couple teams with longer odds a little more than prediction markets, giving Colombia about twice as good odds to win it all, and had been giving Norway good odds until their bracket put them in the same R16 section as Brazil and Japan and the same QF as the winner of Mexico/Ecuador/England/DRC. PELE likes other teams, like England and Portugal, less than prediction markets, conversely.

The reigning champs, Argentina, have a great path to the Semifinals, matching up against Cape Verde in the first round, the winner of Australia/Egypt in the second, and then likely Colombia or Switzerland in the QF. The two next-best teams in the tournament, France and Spain, are both on the far side of the bracket. For this reason, Argentina has consistently risen in the rankings over the last week as the shape of the bracket has become clear:

The United States also has a decent path, with perhaps the easiest first round matchup imaginable in Bosnia and Herzegovina (no offense meant to our Bosnian users).

If the US advances, they’d expect to face the winner of Belgium and Senegal in Seattle, which all things considered is about as good of a corner of the bracket as they could hope to find themselves in. The US have decent odds of a QF appearance in sunny Los Angeles, very likely against a slightly underperforming Spanish squad.

You can also bet on tournament awards, like the Golden Boot (top scorer with assists as a tiebreaker), where Messi is already a heavy favorite due to being up two goals over the next highest candidates after the scoring-heavy group stage.

The Golden Glove award goes to the best goalkeeper, as selected by FIFA’s “Technical Study Group”). Argentinian fan favorite from the previous world cup, Emi Martinez, leads there as well.

Finally, the Golden Ball is voted on from a shortlist by media figures, and is essentially the “Finals MVP” of the World Cup.

Unlike on other prediction market platforms, and unlike Elo-based tournament models… on Manifold you can bet on the real issues at the World Cup.

Traders view it as highly likely that we’ll see more ships transit the Strait of Hormuz than goals at the world cup, although this wasn’t the case just a few days ago.

While initially traders thought some sort of political interference in the World Cup schedule was plausible, it now appears very unlikely over the remainder of the tournament.

One enterprising Manifold user is going to try to get into the World Cup Final in New Jersey for under $500:

User “nsokolsky” has written up the following criteria:

Resolves YES if all hold:

• I, or a friend acting as designated proxy, arrive at a matchday access point with a notable concentration of ticket holders bound for the final (Penn Station shuttle queue, Secaucus Junction matchday platform, American Dream Mall walkway, the stadium gates or any equivalent gathering point), no earlier than 1:30 PM ET on July 19 2026, holding no valid ticket

• Between arrival and the start of the second half, I/proxy obtain a ticket through any means: cash to a stranger, guest-entry on someone else’s account, credential transfer, last-minute FIFA release, FIFA Resale Marketplace, StubHub/SeatGeek/Vivid Seats or any other online platform

• Total cash out the door is strictly under $500 USD. This includes ticket price plus all fees, taxes and transfer charges. Foreign currency converts at the spot rate at time of purchase. Free gifts from strangers count as $0

• The QR scans at every required checkpoint and I/proxy walks into the stadium before the start of the second half

• Any seat is fine. Nosebleed, obstructed view, standing room, whatever

Unlike “nsokoslky,” it’s almost guaranteed that another famous American will be in attendance:

Finally, in a surprisingly uncertain market, one creator has given handicaps to teams with worse odds, in a market on which team will score the most goals over the entirety of the World Cup:

Happy Forecasting!

-Above the Fold

The Trump administration believed them! Or at least, someone did. On June 12th, three days after the launch, Commerce Secretary Howard Lutnick sent a letter to Dario Amodei directing Anthropic to suspend access to both models for all foreign nationals, citing national security authorities. The stated reason was a jailbreak that could unlock the cybersecurity capabilities Anthropic had been carefully routing around.

The tip came from an unnamed competitor. The notice was verbal only. No written technical evidence. Anthropic has no nationality verification system, so an order targeting foreign nationals produced, in practice, a global shutdown for everyone.

By that evening, Claude’s landing page read: Fable 5 is temporarily unavailable. This appears to be the first time a government has forced a publicly deployed frontier AI model offline.

The consistency problem is the part I keep coming back to. GPT-5.5, which Anthropic says has the same cybersecurity capabilities, was not touched. The DoD had already classified Anthropic as a supply chain risk to national security in February (a label usually reserved for Huawei), after Anthropic refused to let Claude be used for mass domestic surveillance and fully autonomous weapons. Anthropic filed suit.

Trump later wrote on Truth Social:

Anthropic said no amount of intimidation would change its position. The Fable 5 shutdown arrived five months into that fight, triggered by a tip from an unnamed competitor whose identity nobody has published. Someone should probably ask who.

Anthropic said it plainly:

We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people. If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.

I think this is right. Every frontier model can be jailbroken given sufficient effort. Any competitor can report a jailbreak to a Commerce official. If claimed jailbreaks are sufficient grounds for recall, you have handed a veto over AI development to anyone willing to run a technique and make a phone call. Whether the underlying capability concern is legitimate (to be fair, Fable did seem kinda scary with the little time I had with it) does not change much about how the mechanism here looks.

There is also quite a bit of irony to the classifier situation. Fable 5 launched with classifiers that silently rerouted suspicious queries to Opus 4.8, the older model that remains available. Users almost immediately noticed the classifier triggering on routine greetings. Anthropic apologized and adjusted it on June 11th. The government issued the shutdown order the following evening. Anthropic built safety infrastructure into Fable 5 specifically because Mythos is dangerous enough to require it, and that safety infrastructure publicly quantified just how dangerous Mythos was.

Onto the markets! There are a lot of them.

The most liquid is “Will Fable be reenabled for Americans before June 20?” at 17%

“Will Fable be reenabled for Europeans before July 1?” has moved dramatically since Anthropic’s Managing Director of International Chris Ciauri told a Seoul press conference on Wednesday that they were “very confident that in the coming days, the models will become available again.” That market has gone from 22% to 57%, up 35 points.

The prop bets market on how restoration looks has also shifted. “Fable available with safeguards more stringent than previously, even for Americans” leads at 50%, down from 57%. The two-tier scenario is at 17%.

Separately, someone has created a market asking whether export controls on Fable 5 will be rescinded within 72 hours of GPT 5.6’s release, which sits at 26%. The competitive pressure theory, in other words, has its own market now.

I should mention that Anthropic had just confidentially filed for an IPO when all of this happened, because there is a theory circulating that deserves at least acknowledging. The narrative the shutdown generates, a model so powerful the US government shut it down for national security, is extraordinarily good IPO marketing.

You cannot buy that kind of capability signal. But IPO pricing is about long-term narrative, not a week of token movement, and the administration is so afraid of Claude that they forced it offline is a better founding story than almost anything Anthropic could have written themselves. Whether the shutdown was anticipated, welcomed, or genuinely unwelcome is something only Anthropic knows, but the incentive structure is at least worth being aware of.

What comes after Mythos in the model naming sequence? Haiku, Sonnet, Opus, Fable, Mythos.

There is a market on this, appropriately. Nobody currently knows, and at the moment nobody can access Mythos to ask it.

Happy Forecasting!

- Above the Fold

I want to start with a huge shoutout to @Genzy and @wasabipesto for all of the collaboration, feedback, reviews, and generally just making Manifold a fun place to be (and build). I’ve learned so much from working with our little team, and I love having the freedom and support to experiment here.

Okay, on to the changes.

In this update:

• @ManifoldSports is live — a new official account bringing tournament match markets to Manifold, starting with the 2026 World Cup (first games June 11th).

• Auto-generating, auto-resolving markets — World Cup matchups create themselves as the bracket fills in and resolve to the official result shortly after each match, so you can roll your mana into the next round without waiting on manual resolution.

• Dashboards are back — relaunching with a redesigned 2026 World Cup dashboard, split into an Official tab (Manifold-run match markets with a new card design and in-page betting) and a Community tab (a preview of how dashboards will work site-wide).

• Refreshed Community Guidelines — moved off Notion and onto native site pages, reviewed for accuracy, with a User Guide on the way.

Welcome @ManifoldSports

Okay, @ManifoldSports isn’t a site change, but what can I say, I’m taking some liberties with this piece and slipping it through before anyone says otherwise.

ManifoldSports is our newest official account, bringing official tournament match markets to Manifold. We’re getting started with the World Cup, which kicks off with two games played in Mexico on June 11th. Follow ManifoldSports on the site for updates, announcements, and more.

Auto-Generating and Auto-Resolving Markets

They look the same as other markets on the site, but ManifoldSports World Cup matchups are different from any others we’ve ever created: these markets auto-generate and auto-resolve after the match is played.

When the World Cup starts, 72 matchups (out of 104 total) are already determined. Those games are already up on the site and available for traders to follow (and bet) along. From there, new matches are determined based on the outcomes. ManifoldSports will auto-generate those markets as they are confirmed.

The first 72 matches will resolve either to a winning team or a Draw. Future matches can’t end in a draw, and those markets are generated with only the option to bet on the playing teams. The close time is set with plenty of time for additional gameplay to break the tie.

As you can imagine, manually managing that many markets takes a lot of attention and work. Traders with profits on these games want to be able to take their Mana and put it to work on the next round, so any delays can be really annoying.

These markets will automatically resolve to the official outcome shortly after the game is played.

Dashboards Are Back (Starting With the World Cup)

If you’re newer here, you may not know Manifold once had Dashboards. If you’ve been around a while, you might (like me!) have missed them. Either way, they’re back! They’re limited for now, but it’s a start.

At the moment, the only one is the 2026 World Cup dashboard, and it has two tabs: Official and Community. They look and behave a little differently from each other and from the rest of the site, so here’s what to expect from each.

Official Tab

This tab holds Manifold’s own match markets, all created and managed by ManifoldSports (with support from the Admin and Mod teams), with a brand-new card design built just for the tournament.

Layout: matches are grouped by date, then sorted by kickoff time, which you’ll see in the top-right of each card. Every card shows live probabilities and market volume at a glance.

Your positions: if you hold one, the card footer updates to show it. A solid pip means at least one position; an open pip means at least one open limit order. Hover for the details.

Betting in place: there’s a redesigned betting modal that opens right on the dashboard. From it you can move between the match outcomes, switch between Yes and No, and place a bet or limit order — all without leaving the page.

Or do it the classic way: the markets themselves haven’t changed, so you can click the match title, the “view market →” link, or the position hovercard to go straight to the full market page.

One thing that’s different: played matches don’t pile up. Once a market resolves, it sinks below the open ones — first into a Recent section, then into Past Games after about 12 hours. You can always scroll down to find resolved markets; you just won’t have to push past them to see what’s still live.

Community Tab

The Community tab is meant to feel more familiar, so the cards stay close to what you see elsewhere on the site, with a few tweaks: they’re resized to fit two across, carry a market-type label in the top right corner, show a chart illustration on numeric markets, and display a positions pill in the bottom-right (with a hovercard) when you hold a position or have an open limit order. As elsewhere on the site, the whole card is clickable and takes you to the market page.

Sorting works a little differently on the Community tab. Resolved markets still sink, but they drop straight into a Past section rather than passing through an interim stage. You also get a few sort options, and this tab accepts all market types — with Polls grouped at the top in a section you can collapse to tuck them away. We wanted polls to stay visible without getting lost among, or mixed into, the regular market cards.

This tab is also our model for where Dashboards are headed: it’s the template we plan to refine and then roll out more widely, eventually letting users build their own. So if you’ve got thoughts on the layout or behavior, comment to let me know. You may help shape the future of Dashboards on the site.

Community Guidelines Refresh

With all respect to Notion, it was time for the Community Guidelines to move on to purpler pastures.

We not only reviewed the content to ensure everything there reflects the current state of Manifold, but we’ve officially created pages on the site for all of that information to live.

Now, when you click the Community Guidelines link, you’ll land on a hub for a variety of pages covering all sorts of details on how the site works, what Moderators do, the ins and outs of participating in Prize Drawings, and a lot more.

The next step is to create a User Guide (there’s a loose starting point in the pages right now) that will offer some insights like best practices, advice on creating interesting markets with legible criteria, and that sort of thing. Have something you think belongs in the User Guide, or the Community Guidelines? Something in there that could be more clear? Let me know!

-Stef