Crossbook
 

The 2026 midterms, scored

Crossbook is an experiment: an AI forecasts prediction markets without seeing the market price, and every call is logged before it resolves and scored against the price at that moment. These are its calls on the US general election of 3 Nov 2026, worst against the market first.

The result

On the midterms

The market was ahead.

Brier 0.1686 for Crossbook, 0.0983 for the market, over 12 settled calls. Lower is better. That is too few calls to be evidence on their own.

On the whole record

The market is ahead.

Brier 0.2061 for Crossbook, 0.1673 for the market, over 174 settled calls on every topic.

Was it one bad night? Across the whole record Crossbook trails the market, and the midterm calls account for 12% of that gap. So no: most of it is spread across every other topic it forecasts.

Every midterm call, worst first

Each call shows Crossbook’s probability, the market price at the moment it was logged, and how it scored. A call’s Brier score is the squared distance between its probability and what happened.

  1. Will the Democratic Party win the NC-13 House seat?

    Crossbook 24.8%Market 43.5% (PM, 31 Oct 2026)Resolved YES. Worse than the market

    Brier 0.5663 against the market’s 0.3192

    Republican incumbent Brad Knott won this district by more than 70,000 votes in 2024, a landslide margin that places NC-13 deep in safe-Republican territory;

    t-0291 · on GitHub late, 9 Nov 2026

  2. Will the Republican Party hold below 190 House seats after the 2026 midterm elections?

    Crossbook 22.8%Market 40.5% (PM, 1 Nov 2026)Resolved YES. Worse than the market

    Brier 0.5968 against the market’s 0.3540

    The Republican Party, as the president's party, enters the midterm with a historically high probability of losing seats (18 out of 20 midterms).

    t-0294 · on GitHub late, 9 Nov 2026

  3. Will the Democrats win the Maine Senate race in 2026?

    Crossbook 55.0%Market 67.5% (PM, 25 Sept 2026)Resolved YES. Worse than the market

    Brier 0.2025 against the market’s 0.1056

    The base rate heavily favors incumbent Susan Collins, but the only direct case-specific evidence — a September 14-15 Quantus poll showing Collins 48, Jackson 47 — is a near-tie, which dominates the incumbency prior and p…

    t-0216 · on GitHub at publication

  4. Ohio Governor winner?: Vivek Ramaswamy

    Crossbook 32.0%Market 20.0% (KS, 29 Oct 2026)Resolved NO. Worse than the market

    Brier 0.1024 against the market’s 0.0400

    Ohio has trended Republican in recent presidential and Senate elections (Trump won by 8+ points in 2020), giving the GOP a structural advantage that suggests a base rate above 50% for the Republican nominee in an open-se…

    t-0283 · on GitHub late, 9 Nov 2026

  5. Will the Democrats win the Ohio Senate race in 2026?

    Crossbook 63.7%Market 73.5% (PM, 1 Nov 2026)Resolved YES. Worse than the market

    Brier 0.1314 against the market’s 0.0702

    The base rate from the last five Ohio Senate elections gives Democrats a 40% historical win rate in the state.

    t-0293 · on GitHub late, 9 Nov 2026

  6. Billionaire one-time wealth tax passes in California election 2026?

    Crossbook 32.8%Market 22.5% (PM, 30 Oct 2026)Resolved NO. Worse than the market

    Brier 0.1073 against the market’s 0.0506

    The qualifying condition is fully satisfied:

    t-0286 · on GitHub late, 9 Nov 2026

  7. Will the Republicans win the Texas Senate race in 2026?

    Crossbook 36.8%Market 31.5% (PM, 29 Oct 2026)Resolved NO. Worse than the market

    Brier 0.1351 against the market’s 0.0992

    The only direct evidence is a Texas Southern University poll (Oct 21-25) showing Democrat Talarico leading Republican Paxton 49-47 among likely voters five days out;

    t-0284 · on GitHub late, 9 Nov 2026

  8. 2026 Balance of Power: R Senate, R House

    Crossbook 19.3%Market 7.5% (PM, 25 Oct 2026)Resolved NO. Worse than the market

    Brier 0.0371 against the market’s 0.0056

    The base rate already encodes the midterm penalty:

    t-0272 · on GitHub late, 9 Nov 2026

  9. Will the Republican Party win the NC-11 House seat?

    Crossbook 21.8%Market 13.5% (PM, 30 Oct 2026)Resolved NO. Worse than the market

    Brier 0.0473 against the market’s 0.0182

    The base rate favors Republicans:

    t-0287 · on GitHub late, 9 Nov 2026

  10. Will the Democratic Party control the House after the 2026 Midterm elections?

    Crossbook 83.3%Market 92.5% (PM, 25 Oct 2026)Resolved YES. Worse than the market

    Brier 0.0281 against the market’s 0.0056

    The base rate strongly favors Democrats:

    t-0274 · on GitHub late, 9 Nov 2026

  11. Will the Republicans win the Nebraska Senate race in 2026?

    Crossbook 82.8%Market 78.5% (PM, 31 Oct 2026)Resolved YES. Better than the market

    Brier 0.0298 against the market’s 0.0462

    Republican incumbent Pete Ricketts is the nominee of the state's dominant party;

    t-0289 · on GitHub late, 9 Nov 2026

  12. Will the Republicans win the Michigan Senate race in 2026?

    Crossbook 19.8%Market 25.5% (PM, 29 Oct 2026)Resolved NO. Better than the market

    Brier 0.0390 against the market’s 0.0650

    The dominant signal is the Oct.

    t-0285 · on GitHub late, 9 Nov 2026

  13. Will Republicans lose the Alaska, Iowa, Kansas, Maine, Michigan, Ohio, and Texas Senate seats?

    Crossbook 13.8%Market 22.5% (PM, 31 Oct 2026)Not yet settled

    The YES outcome requires Republicans to lose all seven seats;

    t-0290 · on GitHub late, 9 Nov 2026

1not yet settled. A human marks every outcome by hand against the venue’s settlement, and each joins this page when it does.

What the timestamps prove

Every call is also written to a public GitHub ledger, whose commit dates Crossbook cannot change. That is meant to prove each call existed before the result was known. For 12 of these calls it does not: a bug meant they were not committed when they were published, and they were added on 9 November 2026, after the election, in a commit marked late:. For those, the site’s own date is the only evidence. The other 1 was committed at publication. Each line above says which.

Calls on GitHub that are not scored

3 more midterm calls were made but held before release, so they never entered the record and count toward nothing. The same bug described above committed them to the GitHub ledger anyway, so they are public there. Here they are, so nothing on GitHub goes unexplained:

  • t-0204 · Will Democrats sweep the Maine Senate and Governor elections? · Crossbook 40.0%, market 69.5% (20 Sept 2026)
  • t-0206 · Will the Republican Party control the Senate after the 2026 Midterm elections? · Crossbook 73.0%, market 38.5% (20 Sept 2026)
  • t-0207 · Will the Democratic Party control the Senate after the 2026 Midterm elections? · Crossbook 19.0%, market 60.5% (20 Sept 2026)

Which calls are here

A rule, not a list. Every published call logged between 1 Aug 2026 and the opening of the polls on 3 Nov 2026 whose market is about Senate or House control, a Senate, House or governor race, or a measure on the 3 Nov 2026 ballot. Primaries, nominations and other countries’ elections are left out. Settled calls count only if the record counts them, by the same rules as /record.

Nothing about how Crossbook forecasts changes because of this result. A record that rewrites its rules after a bad night is not worth keeping. The method is at /methodology and the rules at /rules.

Research and education only. Not investment advice. Not affiliated with Kalshi or Polymarket.