AFOS Analytics
Real-time political pricing
Prediction markets × polls × news
no smoothed averages
United StatesIn the week of August 31 to September 4 the Republican side of both control contracts did not move a single time. The Republican House contract stayed at 11.50% across all 30 readings of the week and the Republican Senate contract stayed at 49.50% across all 29, with a range of 0.00pp in both. This is not an absence of trading: the two took in USD 63,454 of new money over the five sessions. It is a price the market reaffirmed, not a price nobody touched.
That changes the meaning of what moved on the other side. The Democratic Senate contract rose 1.00pp, from 50.50% to 51.50% at the Sep 2 close, and stayed there. But with the Republican leg pinned at 49.50%, that point did not come from anywhere: it went into the sum of the pair, which opened from 100.00% to 101.00% and stayed open through the last three sessions. Normalizing the pair to 100, the Democratic price goes from 50.50% to 50.99%. The move is worth +0.49pp, not +1.00pp. Half of what the raw number says is the book widening.
The House repeated the same thing in miniature. On Sep 3 the Democratic side went to 89.50% with the Republican still at 11.50%, and the pair opened to 101.00%. Raw, +1.00pp. Normalized, +0.11pp. At the Sep 4 close the contract was back at 88.50% and the pair back at 100.00%.
In the Senate seat distribution the mechanics were the opposite, and that is why it is the finding of the week. The sum of the eleven bands is identical at both closes, 98.10%, and only three bands moved. A full point migrated from the band of 53 Republican seats, which is the number they hold today, to the band of 47 or fewer, which is the worst outcome on the table for them. From the best extreme to the worst, crossing the entire width. It did so on USD 3,902 of new money in the whole week, and the band that gave up 0.80pp took in USD 6.
The effect is that the two prices for the same outcome drifted apart: the distance between the Republican binary, at 49.50%, and the normalized distribution, at 45.46%, opened from 3.02pp to 4.04pp at the daily closes, and it opened entirely because the distribution moved while the binary did not.
The week had a ballot box, and it was a Senate one: the Massachusetts primary was on Sep 1. The Senate binary moved at the Sep 2 close, one day later. AFOS records both dates and asserts no link between them. The polling base, for its part, did not update: the most recent field date in the window is Aug 31, and no round has field dates inside Sep 1 to 4.
1.Executive Summary
For a desk, the summary of the week is that almost nothing was repriced in the American Senate and its two books appear to say otherwise, each in its own way.
The binary records a Democratic gain of 1.00pp that did not come from the Republican side, immobile at 49.50% across 29 readings. It came from the pair no longer summing to 100.00%. Whoever reads the raw number sees a point; whoever normalizes sees barely half of one. The difference is not a matter of method preference: it is the difference between a point somebody paid for and a point that appeared because both legs stopped closing.
The seat distribution records a move of similar size and the opposite signature. The sum of the eleven bands is the same at both closes, 98.10%, so every point that came in went out of another band. That is a transfer, and a transfer is information. What weighs is its price: USD 3,902 of new money over five sessions, in a contract carrying USD 2.87M accumulated.
Where there was real money, the price did not move. Where the price moved, the money was marginal.
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 two magnitudes below do not share a unit, and the difference between them is not a number with meaning.
<strong>Why it matters:</strong> 51.50 minus 5.14 is 46.36, and that number does not exist. It has no unit, it would change size if the poll were expressed on another scale, and it corresponds to nothing that can be checked on November 3. Here the mismatch is even larger than in the House: the generic ballot asks about the vote for CONGRESS in the respondent's district, and the Senate has no equivalent national question, because only a third of the seats are up and each one is a statewide race.
The cleanest example of this sits inside the market itself. The binary pays 11.50% on Republican control of the House. Its seat distribution can neither confirm nor deny that number, because the band of 215 to 219 seats contains the 218 that decides control. Adding only the bands that guarantee a majority, from 220 up, gives 9.70%; adding the band that straddles the threshold as well gives 16.70%. The binary fits anywhere in that interval, and the interval widened during the week, from 5.20pp to 7.00pp, because 1.80pp migrated precisely into the band that decides nothing. That is a fact about how the contract was cut, permanent and independent of the election. It is also why the Senate is the only place on this panel where the two books can be placed side by side: there the threshold falls on the edge between the band of 50 and the band of 51.
3.Weighted scenarios for the week
The three scenarios below are not narrative: they are a direct reading of the Senate seat distribution, the contract in which the market publishes the probability of each band. The values are from the Sep 4 close, the last session of the covered week, and the eleven bands sum to 98.10%, inside the 95 to 105 gate AFOS requires to publish a distribution. The three scenarios cover that entire sum, with no overlap and no gap. The percentages are the market's, not the writer's, and do not constitute a recommendation.
The Republicans end with 49 seats or fewer. It is the sum of three bands: 26.50% at 47 or fewer, 13.50% at exactly 48 and 13.50% at exactly 49. This aggregate gained 1.00pp during the week, and the gain sits entirely in the most extreme band, the one at 47 or fewer, while the other two stayed put. In this case the Senate changes hands and the House, which the market already gives to the Democrats at 88.50%, follows.
The Republicans end with 51 seats or more and govern without depending on the tiebreak. It is the sum of seven bands, from the 51-seat band at 10.50% to the 57-or-more band at 0.90%. The aggregate gave up 1.00pp during the week, and the loss came from two bands only: 0.80pp from the exactly 53 seats band, which is the number they hold today, and 0.20pp from the 55 band. The other five did not move.
The Republicans end with exactly 50 seats and the tiebreak goes to the vice presidency. This is the scenario that makes the binary and the distribution disagree, because it counts as Republican control in the binary and appears as an isolated band in the distribution. It stayed put during the week, at 12.50% at both closes. More than one in four Republican control scenarios runs through here rather than through an outright majority: 12.50 out of 44.60.
4.Indicator Grid
| Contract | Current | Δ week | Vol USD acc. | Implied reading |
|---|---|---|---|---|
| House, Democratic control | 88.50% | 0.00pp | USD 5.74M | Third consecutive week closing where it opened. Rose to 89.50% on Sep 3 and came back, with USD 96,064 of new money, the largest flow on the grid. |
| House, Republican control | 11.50% | 0.00pp | USD 4.34M | A range of 0.00pp across the 30 readings of the week, with USD 39,642 of new money. A price reaffirmed under trading, not a price forgotten. |
| Senate, Democratic control | 51.50% | +1.00pp | USD 2.45M | Rose at the Sep 2 close and stayed. Normalized by the pair, at 101.00%, the gain is 0.49pp. |
| Senate, Republican control | 49.50% | 0.00pp | USD 1.68M | A range of 0.00pp across the 29 readings, with USD 23,812 of new money. It is the pinned leg that makes the pair open when the other one moves. |
| Senate, 47 Republican seats or fewer | 26.50% | +1.00pp | USD 0.18M | The most probable band in the distribution and the only one that gained during the week, absorbing the entire point that left the Republican side on USD 2,695 of new money. |
| Senate, exactly 53 Republican seats | 4.50% | -0.80pp | USD 0.09M | Largest band drop of the week, for the second issue in a row. It is the scenario in which the Republicans neither lose nor gain a single seat, and it took in USD 6 across five sessions. |
| Senate, 50-50 tie | 12.50% | 0.00pp | USD 0.13M | The band the binary counts as Republican control and the distribution isolates. Flat during the week, it is the arithmetic source of the divergence between the two books. |
| House, 215 to 219 Republican seats | 7.00% | +1.80pp | USD 0.01M | Largest band gain on the entire grid. It is the band that contains the 218 that decides control of the House, so probability entering it does not resolve the question, it widens the answer. |
| The election happens as scheduled | 96.80% | +0.70pp | USD 0.40M | Largest move of the week among the binaries, within a range of 0.80pp across 31 readings. A small contract by volume and the steadiest on the grid by level. |
5.Liquidity and market structure
The fifth row is the anomaly of the grid and it has been repeating since July. The band of exactly 54 Republican seats is the most traded of the entire Senate distribution, with USD 0.78M accumulated, and the market prices it at 4.00%. The band of 47 seats or fewer, which is the most probable on the table at 26.50%, carries USD 0.18M, four times less. High volume with low probability is concentrated conviction being priced, and it is worth reading alongside this week's figure: the eleven bands together took in USD 3,902 of new money across five sessions, on USD 2.87M accumulated. The 1.00pp displacement described in this issue happened on that flow, and the size of the book is part of reading the number.
6.Electoral calendar through November
| Date | Sample | Why it matters | |
|---|---|---|---|
| Tue, Sep 8 | New Hampshire primary | 1 Senate seat, OPEN | The only open seat among the four September primaries: Jeanne Shaheen announced she is not seeking reelection. A seat without an incumbent is the one that moves the Senate seat distribution the most, which is the contract that moved this week. |
| Wed, Sep 9 | Rhode Island primary | 1 Senate seat | Jack Reed is seeking reelection. A second consecutive day of voting, and the state votes on a Wednesday, outside the Tuesday pattern of the rest of the calendar. |
| Tue, Sep 15 | Delaware closes the primary calendar | 1 Senate seat, last state | Chris Coons's seat and the last primary of the cycle. From there the candidate field is closed in every state, and the market starts pricing a settled contest. |
| Nov 3 | General election | 435 House seats and 1/3 of the Senate | The date on which every contract on this grid resolves. Until then AFOS measures the distance between the price and the yardstick. |
There is no mandatory poll registry in the United States, so this calendar is one of ELECTORAL EVENTS and not of publication dates. The cadence of the pollsters is each house's decision. The Massachusetts primary, which opened this list in the previous issue, was held on Sep 1, inside the covered week, and leaves here. The Senate binary contract moved at the following day's close, Sep 2. Both dates are recorded and AFOS asserts no link between them.
7.Watch list, week triggers
- The Republican leg of both control contracts, immobile at 11.50% and 49.50% is the central figure of this issue and the one that most deserves watching. That is 59 readings across the two contracts, five sessions, a range of 0.00pp in both, with USD 63,454 of new money. As long as it stays pinned, any variation on the Democratic side will show up inflated in the raw number, because it goes into the sum of the pair instead of coming out of the counterparty.
- The sum of the Senate binary pair, open at 101.00% since Sep 2 is what converts 1.00pp raw into 0.49pp normalized. A pair at 101.00% is not news by itself and usually closes on its own. What is worth watching is the duration: three consecutive sessions open is a sign that one of the legs stopped, not that both are trading.
- The distance between the two Senate books, at 4.04pp at Friday's close opened from 3.02pp during the week, and from one side only: the Republican binary did not move and the normalized distribution fell from 46.48% to 45.46%. If next week repeats the pattern, the distance becomes a trend rather than an oscillation.
- The band of 215 to 219 Republican House seats, at 7.00% gained 1.80pp, the largest band gain on the grid, and it is the one containing the 218 of control. Every point entering it widens the interval in which the binary can be right, rather than narrowing it.
- The popular vote margin contract, structurally outside the gate its fourteen bands sum to 149.60% in the Sep 6 reading, because the contract includes a bucket capturing any other outcome and it is counted in. That is how it was built, not a one-day oscillation. It is the only market that would measure the same magnitude as the poll, and that is why it is collected every day anyway, to build the series. It is never published as if it counted.
8.Methodology
AFOS reports the three signals separately: prediction market price, generic ballot polls with sample size and field dates, and press coverage from a fixed list of outlets. It builds no composite index and does not subtract one magnitude from another.
Every published price passes through two independent readings eight minutes apart, and only ships when they agree within 0.20pp. This issue uses the Sep 6 reading, 19:04 BRT, with an approved verdict. The weekly variations use the CLOSE of each day, which is the last point recorded that day, and not the first and last point of the window, because the second choice changes the number without changing the fact. When the text says a contract had a range of 0.00pp, that is measured across every reading recorded during the week, and the count of those readings is stated alongside. The House series begins on Jul 29, 2026 and the Senate series on Apr 14, 2026, so any superlative about them is always 'of the series', with the start date, and never 'of the cycle'.
When the two sides of a binary contract do not sum to 100, this brief publishes the raw and the normalized value side by side, because the difference between them is exactly what distinguishes a point that changed hands from a point that appeared in the sum. A distribution is only published if its bands sum to between 95% and 105%. In this issue the Senate distribution sums to 98.10% and the House one to 100.60%, both inside the gate, and the three scenarios in section 3 cover the entire sum of the first. The generic ballot average is a simple arithmetic mean, unweighted and with no pollster excluded; when the same house publishes more than one screen of the same round, only one enters, by the likely voter, registered voter, adults hierarchy. Accumulated volume in USD is cited as context and is not guidance; book depth is not cited, by editorial decision of May 21, 2026. Code and method are open under Apache 2.0.
9.Additional reading · macro coverage
AFOS is the primary source of this analysis. The references below are complementary context reading and do not support any of the measurements above.
- The Washington Post · Guide to the most relevant House races in November, useful for mapping where the seats in this grid's distribution would be decided., www.washingtonpost.com/politics/2026/09/06/guide-key-battleground-house-races-this-november/
- The Guardian · Overview of the start of the final stretch for control of Congress, published on Sep 6, the weekend following the week covered by this issue., www.theguardian.com/us-news/2026/sep/06/midterm-elections-control-of-congress
- Bolts · Guide to the September 2026 elections, with the primaries that close the cycle's calendar and what is at stake in each state., boltsmag.org/whats-on-the-ballot/september-2026-elections-guide/
- The Associated Press · Ongoing coverage of the November races and of primary results as the states vote., apnews.com/hub/us-elections
- The Hill · Coverage of the congressional campaigns, including the New Hampshire Senate race, which has its primary on Sep 8., thehill.com/homenews/campaign
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.