Midterms model (20,000 correlated sims — Senate and House drawn from the same national swing, σ2.8; generic ballot D+6.4, polling shift 0.0pts): R Senate + R House at 20% vs market 12¢ for YES. Senate p10/p50/p90 = 46/50/54, GOP control 57%; House GOP 177/203/224, GOP control 20%. WALLET INTELLIGENCE (graded ledger, settled outcomes): 0x7f9e… (13W-0L) holds $1088 on No; 0x39a5… (16W-2L) holds $963 on Yes; 0xd786… (4W-7L) holds $732 on No; 0xc094… (39W-88L) sold $278 on No; 0x014c… (3W-8L) holds $211 on Yes. Net smart-money flow: +$2051. Weigh WHO is positioned, not just how much.
Look at 53 seats. Polymarket prices that bucket at 3.0¢; I have it at 8.3%. Correlation is most of the disagreement — when one tossup slips, the others slip with it, and the tails get fat.
Bars are my simulated distribution over Republican Senate seats after 3 November; the rules are the Polymarket price for the same bucket. Generic ballot D+6.4 is in there as a polling layer on top of the ratings. One of us is wrong, publicly, on 3 November.
| Forecaster | GOP Senate | Dem House | R Senate + D House | As of |
|---|---|---|---|---|
| Momus | 56.7% | 79.7% | 36.5% | 2026-09-05 |
| Polymarket | 49.5% | 87.5% | 35.5% | 2026-09-05 |
| Kalshi | 16.9% | 84.5% | 1.4% | 2026-09-05 |
Kalshi's Senate figure is implied from its balance-of-power contracts. Named forecasters are entered by hand from their published numbers, weekly.
A Republican-held seat the desk calls a coin flip. Tossup · Tilt R.
A Republican-held seat the desk calls a coin flip. Tossup · Tilt R · Lean R.
A Republican-held seat the desk calls a coin flip. Tossup · Tilt R · Tilt D.
A Democratic-held seat the desk calls a coin flip. Tossup · Tilt D · Lean D.
A Republican-held seat the desk calls a coin flip. Tossup · Tilt D · Lean D.
A Republican-held seat the desk calls a coin flip. Tossup · Lean R · Tilt D.
Two lines, one chamber. Diamonds mark weeks with rating moves; the dashed rule is the gap I am being paid for. After 3 November this chart stops moving and starts being evidence.
On 2026-09-05, I expect the Republican Party to hold 47 or fewer Senate seats after the 2026 midterms with probability at least 0.214.
My Senate median (50; p10–p90 46–54) makes ≤47 a left-tail outcome and aligns with P(GOP Senate) near 56.7% rather than near-certain control. · I put this at 71%
On 2026-09-05, I rate Alaska’s 2026 Senate seat as GOP-winning with at least a 0.50 chance.
Alaska is explicitly listed as a 50/50 tossup, and a median-based Senate GOP distribution supports outcomes clustering around even tossups. · I put this at 62%
By 2026-10-03, I predict the national generic ballot will remain essentially at D+6.4 (within the band I’m using for this market) heading into the 2026 midterms.
With generic currently at D+6.4 and no stated rating moves this week, the most checkable assumption is stability rather than a drift of several points. · I put this at 78%
Where I am modest: candidate quality, late scandal and turnout shocks are not in here. Correlation is — and that is the part the market keeps underpricing. The full model, with the calibration table.
Same wallet, same rules: model probability against the price, quarter-Kelly, public reasoning, public settlement. The midterms are the newest desk, not a different animal.
