AFOS Analytics
Real-time political pricing
Prediction markets × polls × news
no smoothed averages
United StatesThe week of 21 to 25 September separated two behaviours that are usually read as one. In the House, the binary moved 3.00pp over two sessions and then stopped: the Democratic price closed at 92.50% on Monday the 20th and repeated 92.50% at every subsequent close through Friday, without a single tick. In the Senate, the same kind of contract moved 6.00pp through Wednesday the 23rd, touched 66.50% and gave back 4.00pp, closing the week at 62.50%. The closing delta against the previous Friday is +2.00pp, and it hides the entire round trip. One book moved and stayed; the other moved and came back.
The second finding sits inside the Senate, and it inverts last issue's. There are two ways to buy the same outcome: the control binary and the sum of the ranges in the seat distribution. Last week those two readings converged, and the convergence came almost entirely from the binary travelling to where the distribution already was. This week they moved apart again, and the movement switched sides. The Republican binary shed 3.00pp, from 40.50% to 37.50%. The sum of the ranges that give Republicans a Senate majority shed 9.20pp, from 40.80% to 31.60%, and normalised by the book's own sum it shed 6.79pp, from 39.50% to 32.71%. What moved, this time, was the distribution, and it moved about two and a half times what the binary moved.
The practical effect is a distance of 4.79pp in implied probability, or 5.90pp in the price you pay, between two ways of buying a Republican Senate majority. It exists in books on the same event, at the same close, and it depends on no forecast of ours: it is arithmetic over published prices. For a trading desk, it is a distance between two instruments, and it is the most concrete measurement in this issue.
1.Executive Summary
For a desk, the week reads as stability in the House and growing disagreement between the two Senate books. The House binary repriced 3.00pp and then froze: six closes at the same value, with the pair summing to exactly 100.00% at both ends, which means there is no overpricing to discount in that number. The Senate is the opposite: the binary made a 6.00pp run to the top of the series and gave back two thirds of it, and the seat distribution of the same Senate moved almost three times as much in the same direction, opening a 4.79pp distance in implied probability between two ways of buying a Republican majority. The poll does not enter that arithmetic, and the next section explains why it cannot.
2.Why this brief does not subtract the market from the poll
This brief never subtracts the probability of controlling a chamber from a lead in vote points. The two measures answer different questions, and the difference between them has no unit. This issue contains a direct demonstration of that inside the market itself, without having to reach for the poll.
Subtracting 7.57 from 92.50 would produce 84.93, and that number does not exist: it would change size if the lead were expressed as a fraction of the vote instead of percentage points, and it would change sign in historically possible scenarios. In 2012 Democratic House candidates won more votes in the national total and ended with fewer seats, because converting votes into seats depends on geography and district design. The edge between these two numbers, in the dashboard graph, is mute on purpose.
And this week's demonstration does not have to leave the market. Inside the market itself, the two Senate books have different resolutions: 50 seats is a range border in the distribution, so the sum of the Republican-majority ranges is exact. In the House it is not: the 218-seat threshold falls INSIDE the 215 to 219 range, so that distribution cannot be summed into a Republican majority, it can only bracket it. Buying from 220 seats up costs 5.80%; including the range that contains the threshold costs 11.60%. The binary asks 7.50%, and it sits inside the interval, which is consistency and not coincidence. One book answers at five-seat resolution and the other at binary resolution: it is the same lesson as the section above, measured in two contracts on the same event.
3.Weighted scenarios for the week
In the United States the scenario does not have to be invented, because there is a contract that pays by seat range. The three below come from the distribution at the 25 September close, normalised by each book's own sum, which closed at 104.90% in the House and 96.60% in the Senate. None of them is an AFOS forecast: they are price readings.
It is the most expensive block in both books. In the House, the ranges below 200 Republican seats sum to 60.50% gross, or 57.67% normalised, and the single most expensive range in the whole book is the one below 190 seats, at 27.50%. In the Senate, the ranges up to 49 Republican seats sum to 65.00% gross, or 67.29% normalised. Both binaries agree with that reading: 92.50% for Democratic control of the House and 62.50% for the Senate.
The sum of the ranges at 50 Republican seats or more closes at 31.60% gross, or 32.71% normalised. The Republican Senate control binary closes at 37.50%. The distance is 4.79pp in implied probability and 5.90pp in the price you pay, and it OPENED during the week: at the 18 September close the distribution paid 39.50% normalised against the binary's 40.10%, a distance of 0.60pp. The movement came from the distribution, which shed 6.79pp normalised, against the binary's 2.60pp.
The 218-seat threshold falls inside the 215 to 219 range, so the House distribution does not sum that outcome, it bounds it. The floor is 5.80%, buying from 220 seats up, and the ceiling is 11.60%, including the range that contains the threshold. The binary asks 7.50%, inside the interval. The book's two highest ranges, 225 to 229 and 230 or more, close at 1.10% and 0.60% and lost 1.80pp and 0.00pp on the week.
4.Indicator Grid
| Contract | Current | Δ week | Vol USD acc. | Implied reading |
|---|---|---|---|---|
| House, Democratic control | 92.50% | +3.00pp | USD 7.07M | Closed at 92.50% on the 20th and repeated at six consecutive closes. Pair at 100.00% at both borders, so raw and normalised coincide. |
| House, Republican control | 7.50% | -3.00pp | USD 5.22M | Intraday floor for the week at 6.50%, reached on 20 September at 11:30 UTC, which is also the floor of the series opened on 29 July. |
| Senate, Democratic control | 62.50% | +2.00pp | USD 3.14M | Touched 66.50% at Wednesday's close, the top of the series opened on 14 April, and gave back 4.00pp by Friday. Intraday amplitude of 8.00pp on the week. |
| Senate, Republican control | 37.50% | -3.00pp | USD 2.13M | Intraday floor at 34.50%, on 22 September at 22:00 UTC. The pair deflated from 101.00% to 100.00%, so the normalised loss is 2.60pp and not 3.00pp. |
| Republican Senate, sum of the ranges at 50 seats or more | 31.60% | -9.20pp | USD 3.01M | The same question as the Republican Senate binary, bought by range. Normalised by the book's sum, which closed at 96.60%, it gives 32.71%, against the binary's 37.50%. |
| Election happens as scheduled | 98.20% | +0.10pp | USD 0.45M | Intraday amplitude of 0.70pp on the week, with a floor at 97.70% and a top at 98.40%. It is the quietest of the five books this issue cites. |
5.Liquidity and market structure
6.Electoral calendar through November
| Date | Sample | Why it matters | |
|---|---|---|---|
| 23/Sep, Wednesday | The Missouri map returns to the Supreme Court for the third time | 8 House districts | District redesign is the only variable in this grid that changes the conversion of votes into seats without any voter changing their mind. It is exactly why this brief does not subtract a poll from a control probability. |
| 24/Sep, Thursday | Voter-registration backlog in Texas | a state with 38 seats | It enters the grid as an administrative fact with a deadline, not as an effect estimate. AFOS does not convert this into seats or into vote points. |
| 03/Nov, Tuesday | Election day | 435 House seats and 35 Senate seats | The calendar contract closes the week at 98.20%, with an intraday amplitude of 0.70pp. There are 39 days to go, counted from the 25 September close. |
There is no mandatory poll registration in the United States, so this grid does not anticipate release dates: it lists what is already scheduled and what has already happened with a date verified at the outlet. The congressional primary season closed on 15 September, and from here to November what moves price is polling, campaign flow and court decisions on maps.
7.Watch list, week triggers
- The distance between the two Senate books It closed to 0.60pp on 18 September and opened to 4.79pp on the 25th. If it closes again, what matters is which side moves: last week it was the binary travelling to the distribution, this week it was the distribution moving away from the binary.
- The freeze in the House binary Six closes at 92.50%, with the pair summing to exactly 100.00%. While the pair stays at 100.00%, raw and normalised coincide and any move in the price is a move in the implied probability, with no spread to discount.
- The range below 190 Republican seats It is the single most expensive range in the House book, at 27.50%, and together with the 190 to 194 range it gained 5.00pp on the week. It is where the distribution is most concentrated.
- The fieldwork date of the polling average The 30-day window has its most recent fieldwork on 22 September. When the fieldwork date stops advancing, the average can move without any voter being asked again, because an old round drops off the edge: in that case the variation is window composition and does not read as a move in opinion.
8.Methodology
Polymarket prices, read through the AFOS proxy with a live and uncached reading, and confirmed by a double reading eight minutes apart with a 0.20pp tolerance before any number goes live. A daily close is the last point recorded that day in the series backup, not the API reading, and the floors and tops quoted as intraday come from the 122 readings per range and the 50 to 58 readings per binary recorded during the week. The weekly delta is measured against the PREVIOUS FRIDAY's close, 18 September, which is this series' convention. A binary pair is reported raw and normalised by the sum of the two outcomes, and that sum is declared at both borders whenever it changes. A seat distribution is only cited when the ranges sum to between 95% and 105%: in this issue the House closed at 104.90% and the Senate at 96.60%, and any sum of ranges is normalised by its own book's sum before being compared with a binary. The convention for a Republican Senate majority is 50 seats or more, because the Vice Presidency breaks ties, and 50 is a range border in that book, which makes the sum exact. In the House the 218 threshold falls inside a range, so there the distribution is reported as an interval and never as a sum. The generic ballot is a simple arithmetic average, unweighted and with no pollster excluded, over a rolling 30-day window, with one round per pollster per wave by the likely voter, registered voter and adult hierarchy. Decimal point in all three languages.
9.Additional reading · macro coverage
AFOS is the primary source of this analysis. The references below are complementary context and do not support 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.
- The Washington Post · The fight over the Missouri map returns to the Supreme Court for the third time, on 23 September., www.washingtonpost.com/politics/2026/09/23/missouri-congressional-map-fight-returns-supreme-court-third-time/
- Axios · Field reporting pointing to Republicans on track to lose the House majority, on 25 September., www.axios.com/2026/09/25/house-republicans-majority-democrats-battleground
- The New York Times · Five takeaways from a new poll in the New Hampshire Senate race, on 24 September., www.nytimes.com/2026/09/24/us/politics/new-hampshire-senate-poll-pappas-sununu.html
- The Washington Post · The Republican position in the South Carolina Senate seat, on 23 September., www.washingtonpost.com/politics/2026/09/23/gop-hold-south-carolinas-senate-seat-is-slipping/
- NBC News · How NBC own polling averages describe the race for Congress, on 25 September., www.nbcnews.com/politics/2026-election/nbc-news-polling-averages-democrats-enjoy-biggest-lead-yet-race-congre-rcna599378
- The New York Times · A Republican financial edge in the final stretch, on 21 September., www.nytimes.com/2026/09/21/us/politics/republicans-midterm-fundraising-maga-inc-trump.html
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.