AFOS Analytics
Real-time political pricing
Prediction markets × polls × news
no smoothed averages
United StatesThe binary Senate control contract paid 60.50% to the Democrats at Friday's close, against 52.50% at the previous Friday's close. That is 8.00pp in one week, and it is the largest weekly shift of the 22 weeks in the series opened on 14 April, above the 7.00pp of the week of 14 August. The move is not book overpricing: the pair summed to 101.00% at both ends, so the delta survives normalisation almost intact, +7.92pp. And it was not gradual. Between Monday's close and Tuesday's the price jumped from 53.50% to 58.50%, five points in one session, and the three days that followed added only 2.00pp.
What makes the week unusual is not the size of the move, it is where it went. This brief has tracked since August the fact that the Senate has two prices for the same outcome: the binary control contract and the Republican seat-count contract, which allows the buckets to be summed and the same probability derived. The two readings had been disagreeing, and at the close of 11 September the distance was 8.07pp. At the close of 18 September it is 1.00pp. The convergence came almost entirely from one side: the binary moved 7.92pp normalised and the distribution moved 0.93pp. The binary came to the distribution, not the other way round.
In the House the week was smaller and in the same direction. The Democratic contract closed at 89.50%, up 2.00pp, and the Republican one at 10.50%, down 3.00pp, which is the largest weekly shift of the 7 weeks in the series opened on 29 July for that side. The pair deflated from 101.00% to 100.00%, so here the normalised and the raw diverge: the Democratic gain is worth +2.87pp after normalising and the Republican loss is worth -2.87pp, not the raw 3.00pp. The 89.50% ties the series high, recorded on 30 August, and does not exceed it.
1.Executive Summary
For a desk, the week reads as convergence inside the Senate and continuity in the House. The Senate binary repriced 8.00pp across five sessions, with almost all of the move concentrated in a single day, and ended up paying very nearly what the seat distribution was already paying before it moved. Anyone following both books had, since 11 September, a distance of 8.07pp between two ways of buying the same outcome; that distance closed to 1.00pp without the distribution needing to move. In the House the picture did not change in nature: the Democratic price sits at the series high and the Republican one at its floor, and what the week adds is the largest weekly loss ever recorded on the Republican side of that book.
2.Why this brief does not subtract the market from the poll
This brief publishes two numbers about the same election and never subtracts them. The market pays 89.50% on the probability that the Democrats control the House. The generic ballot polls in the 30-day window give the Democrats an average advantage of 7.00 points of vote. The difference between 89.50 and 7.00 is not a number: it is an operation between quantities that do not subtract.
The contract that would measure the SAME quantity as the poll is the popular vote margin book, and it is structurally outside the coherence gate: its fourteen buckets sum above 150% because the book includes a bucket capturing any other outcome and counts it alongside the rest. It has been that way since it opened, it is not a one-day reading, and that is precisely why the bridge between the two quantities does not exist in the market.
Inside the market itself, however, there is a pair that CAN be compared, and it is this issue's finding. The binary Senate contract and the Republican seat-count contract measure the same outcome in the same unit, because the control threshold falls on a bucket boundary: 50 seats is its own bucket. Adding the tie at 50 to every bucket from 51 upward, the distribution pays 40.80% on Republican control, or 39.50% after normalising the 103.30% sum to 100. The binary pays 40.50%. The distance between the two is 1.00pp, against 8.07pp a week earlier, and below every reading recorded since 23 August, whose narrowest was 1.21pp. ⚠️ In the House that comparison is not possible: the 218-seat threshold falls INSIDE the 215 to 219 bucket, so the distribution only bounds Republican control between 8.50% and 15.40%, and the binary, at 10.50%, fits inside that band without contradicting it.
3.Weighted scenarios for the week
The three scenarios below are not narrative: they are a direct reading of the Republican seat-count contract in the Senate, at the close of 18 September. The eleven buckets sum to 103.30%, inside the 95 to 105 gate, and the percentages are what the market pays, not estimates from this house. 🔑 The rule that decides the arithmetic is not in the data: a 50 to 50 tie gives control to the Republicans through the Vice President's casting vote, so the tie counts on the Republican side.
The Republicans end with 49 seats or fewer. It is the sum of the three lower buckets, and within it the concentration is heavy: on its own, the 47 seats or fewer bucket pays 33.50%, more than the other two combined, and it was the biggest gainer of the week, up 4.00pp. This is the scenario the binary came around to pricing: it pays 60.50% to the Democrats, and the distribution pays 62.50% on an outright majority, before any normalisation. The difference between the two readings of this same outcome fits inside 2.00pp.
It is the sum of the tie at 50 to 50, which pays 11.50%, with the six buckets from 51 seats upward, which total 29.30%. Normalised, Republican control is worth 39.50%. ⭐ The number a desk should take from here is the composition: more than a quarter of the entire Republican scenario runs through a tie and not through a majority, 11.50 of 40.80, and the tie depends on a casting vote rather than on one more seat. The bucket for exactly 52 Republican seats was the biggest mover in the whole grid, up 4.50pp to 8.90%, on USD 1,762 of new money over USD 0.61M accumulated.
It is the sum of the buckets for 53 seats or more, which is exactly the number they hold today. 🏷️ The label matters and the exact form is lose none, because the bucket is 53 OR MORE and writing hold their 53 suggests exactly 53, which is a different claim. The scenario paid 12.80% at the close of 11 September and 14.25% in the reading of 23 August, so it gave ground on both comparisons. The opposite tail, 57 seats or more, pays 0.60% and is the cheapest price in the grid.
4.Indicator Grid
| Contract | Current | Δ week | Vol USD acc. | Implied reading |
|---|---|---|---|---|
| Senate, Democratic control | 60.50% | +8.00pp | USD 2.82M | Largest weekly shift of the 22 weeks in the series. Path across the five sessions: 53.50, 58.50, 58.50, 59.50, 60.50. The series high was recorded at 23:00 UTC on Friday, which is the close itself. Took in USD 306 thousand over the window. |
| Senate, Republican control | 40.50% | -8.00pp | USD 1.88M | Series floor, recorded at 00:30 UTC on Thursday, across 370 points recorded since 14 April. The pair summed 101.00% at both ends, so the normalised loss of 7.92pp preserves nearly all of the raw move. |
| Senate, distance between the binary and the distribution | 1.00pp | -7.07pp | USD 4.70M | Raw Republican binary minus Republican control derived from the buckets and normalised. It was 8.07pp on 11 September. The convergence came from the binary, which moved 7.92pp, and not from the distribution, which moved 0.93pp. |
| House, Democratic control | 89.50% | +2.00pp | USD 6.61M | Ties the series high, from 30 August, and does not exceed it. Normalised, the gain is larger than the raw one, +2.87pp, because the pair deflated from 101.00% to 100.00%. Took in USD 383 thousand, the grid's largest single inflow. |
| House, Republican control | 10.50% | -3.00pp | USD 4.85M | Largest weekly loss on this side across the 7 weeks of the series. It sat still at 13.50% until Wednesday and gave up the 3.00pp over two sessions. Ties the series floor, from 30 August. |
| Senate, tie at 50 to 50 | 11.50% | -0.50pp | USD 0.13M | The bucket that decides control through the Vice President's vote. It accounts for more than a quarter of the entire Republican scenario and gave up half a point on the week, with an intraday low of 11.50% on Friday evening. |
| Senate, exactly 52 Republican seats | 8.90% | +4.50pp | USD 0.61M | Biggest move in the entire grid, in any direction. It touched 9.10% in the early hours of Friday, with a 5.10pp range on the week and only USD 1,762 of new money. |
| The election happens as scheduled | 98.10% | +0.80pp | USD 0.42M | It rose in two small steps and closed at the week's high. It is the grid's most expensive contract and the one that moves least, with a 0.80pp range across five sessions. |
5.Liquidity and market structure
The fifth largest accumulated volume in the whole grid belongs to a contract paying 2.70%. The bucket for exactly 54 Republican Senate seats has USD 0.78M traded since it opened, more than the ENTIRE House seat book, which totals USD 0.30M across ten buckets. The reading is not that the market believes in 54 seats: accumulated volume is trading history since the contract opened, and that bucket was expensive when the Senate map looked different. The week's flow tells a different story from the stock: the two binary books took in USD 1.01M between the closes of 11 and 18 September, and the entire Senate seat distribution took in under USD 12 thousand. The price that moved 8.00pp is the one that received money; the one that stood still is the one that did not.
Volume is accumulated since each contract opened and the values are those of the 18 September close. AFOS does not cite book depth in editorial text, because low liquidity in a continuously arbitraged market does not mean the price is wrong.
6.Electoral calendar through November
| Date | Sample | Why it matters | |
|---|---|---|---|
| 16 Sep, Wednesday | The House wraps up business until after the midterms | 435 seats in play | The House cancelled votes and brought forward the campaign recess. For the grid the effect is one of calendar: the source of legislative events that moved price during the week disappears, and what remains through November is polling and campaign events. |
| 16 Sep, Wednesday | The congressional primary season closes | record departures and defeats | With Delaware on 15 September and the close of the season, there is no primary left to price. From here the seat contracts depend only on district polling and campaign flow. |
| 03 Nov, Tuesday | General election | 435 House seats and 1/3 of the Senate | The date on which every contract in this grid resolves. Until then AFOS measures the distance between the price and the poll, and publishes both without choosing between them. |
There is no mandatory poll registration in the United States, so this calendar is one of ELECTORAL EVENTS and not of release dates. That is the structural difference from Brazil, where the electoral court publishes in advance who will release and when. Here the pollsters' cadence is voluntary, and that is why the house average always declares the most recent field date feeding it, which in this issue is 17 September.
7.Watch list, week triggers
- The distance between the two Senate books, at 1.00pp fell from 8.07pp in one week and landed below every reading recorded since 23 August, whose narrowest was 1.21pp. If it opens up again, the question is which side moves: so far the binary has done all the work of closing and the distribution has stayed nearly still.
- The 5.00pp jump between Monday and Tuesday in the Senate binary concentrates almost two thirds of the week's move in a single session, and the three days that followed added 2.00pp. A step move and a drift move read differently, and the series will say which this was once next week closes.
- The bucket for 47 Republican seats or fewer, at 33.50% rose 4.00pp and on its own pays more than the 48 and 49 buckets combined. It is the book's bottom bucket, so it captures any result below 47, and it concentrates more than half of the entire outright Democratic majority scenario.
- The House pair, back at 100.00% had been at 101.00% and closed the week exact. While it sits at 100, raw and normalised coincide and the published delta is the same on both readings. When it inflates again the difference returns, and it is what made this week's Republican loss worth 2.87pp normalised against 3.00pp raw.
- The generic ballot average, with most recent field date 17 September stands at D+7.00 across 23 rounds from 18 pollsters. The interval between the most favourable round for each side is 15.00 points, and a single one of the 23 puts the Republicans ahead. It is the dispersion, not the average, that describes what the pollsters are measuring.
8.Methodology
AFOS reports the three signals separately: prediction market price, generic ballot polls with declared sample and field dates, and press coverage from a fixed list of outlets. It does not produce a combined number and does not arbitrate between the sources. A market price goes live only after two independent readings eight minutes apart that agree within 0.20pp, contract by contract, and the reference series are measured in the daily backup of the database and not in the public interface, which caps the window at ninety days. Every claim of an extreme declares the window and the start date of the series in which it was checked.
The generic ballot average is simple and unweighted, over a rolling 30-day window, with a sample-frame hierarchy preferring likely voters to registered voters and those to adults. A bucket distribution only reaches the screen when the sum falls between 95% and 105%. The code and the method are open under the Apache 2.0 licence, and the datasets are published with a permanent identifier.
9.Additional reading · macro coverage
AFOS is the primary source for this analysis. The references below are complementary context and do not underpin any of the measurements above. ⛔ None of them is presented as a cause of price movement: this brief does not attribute repricing to an event without establishing the time window.
- NBC News · Supreme Court rulings against the White House and the Republican party on the eve of the midterms, 15 September., www.nbcnews.com/politics/supreme-court/supreme-court-pushes-back-trump-gop-ahead-midterm-elections-rcna595375
- CBS News · The House cancels votes and brings forward the recess until after the election, 16 September., www.cbsnews.com/news/house-cancels-votes-adjourning-midterms/
- The Guardian · More Republicans turn against the war with Iran as the election approaches, 16 September., www.theguardian.com/us-news/2026/sep/16/republicans-iran-war-midterm-elections
- NPR · Review of the close of the primary season, with record departures and incumbent defeats, 16 September., www.npr.org/2026/09/16/nx-s1-5971264/congress-retirement-2026-house-senate-primary-defeat
- The Hill · DDHQ projection on the electoral balance of redistricting, 15 September., thehill.com/homenews/state-watch/6090677-ddhq-forecast-democrats-house-majority/
Get it by email
AFOS Tradeoff is the weekly read on Brazilian political risk, for people who decide on data rather than headlines.
No spam. Unsubscribe anytime.
After signing up you choose whether to receive it in English, Portuguese or Spanish.