AFOS Analytics
Real-time political pricing
Prediction markets × polls × news
no smoothed averages
United StatesThe previous issue closed by naming the band of 215 to 219 Republican House seats and saying what to watch in it: if it got more expensive, the range in which the two House books can agree would widen; if it got cheaper, the range would close and the two books would start charging each other. It got cheaper, from 6.40% to 5.45% of the book's mass, and they charged each other. The binary contract on Democratic control of the House crossed the week of 12 to 16 October without moving a tick: 44 recorded readings, floor equal to top at 90.50%, amplitude of 0.00pp, and the same 90.50% at the close of each of the five trading days. The pair summed to exactly 100.00% at both edges of the window, so here raw and normalised say the same number and there is no overpricing to discount. And it was not a lack of money: USD 2.40M of new money entered the House book during the week, against USD 0.63M in the Senate book, without the price moving a tick. The market's other book on the same question did the opposite. The distribution of Republican House seat bands, which is the only contract that gives the seat dimension, reorganised its own mass on a large scale: the band of 205 to 209 seats went from 11.50% to 28.90% in raw price, which is an advance of 16.99pp of the book's share, and became the most expensive band in the whole of it, ahead of the band of fewer than 190 seats. Mass left both tails at the same time, and the three bands from 220 seats upward gave ground together. Measured on the same basis at both edges of the week, the share of the book sitting below the majority threshold went from 85.26% to 90.65%, and the share from 220 seats upward fell from 8.34% to 3.89%. The binary stood still at 9.50% the entire time. The two books neither add to nor subtract from each other, and this brief does not convert one into the other: the seats book carries permanent overpricing, sums above 100% and is therefore read for its SHAPE and never for its level, while the binary, with the pair at exactly 100.00%, is read directly. What is being asserted is only the contrast in behaviour: one of the two books on the same question reorganised a quarter of its own mass, and the other did not move. In the Senate the reading is the inverse and requires both measures, as it already required in the previous issue for the opposite reason. The contract on Democratic control rose 2.00pp raw, from 62.50% to 64.50%, but the pair sum went from 99.00% to 101.00% over the period, and normalised the advance is 0.73pp. The Republican side is the cleanest proof of this: it closed the week at the same 36.50% at which it opened, 0.00pp raw, and its implied probability gave back 0.73pp. A contract that did not move a tick in 49 readings and whose implied probability retreated three quarters of a point.
1.Executive Summary
The three numbers above are of different natures and do not combine. The first two are contract prices, read live and confirmed by a double reading 8 minutes apart, with a tolerance of 0.20pp. The third is an arithmetic average of published polls, in a 30 day rolling window. This issue measures how each of them behaved in the week of 12 to 16 October, with the delta anchored to the previous Friday's close, 9 October, and it refuses the subtraction between them.
2.Why this brief does not subtract the market from the poll
The question comes up in every trading desk conversation, and this week it comes up with one side frozen: if the market pays 90.50% for Democratic control of the House, and that price did not move a tick across five trading days, while the poll average gives a Democratic advantage of 7.54 points, what is the distance between the two readings? The answer is still that the question has no numerical answer, and the reason is one of units.
Subtracting 7.54 from 90.50 would produce the number 82.96, which is not a probability, is not a vote point and would change in size if the poll were expressed on another scale, for example as a share of the two party vote. The 2012 case is the short historical example: Democrats won more House votes and fewer seats, because converting votes into seats depends on geography and on district design. This brief reports the two readings side by side and never the difference between them.
There is an instructive comparison this week, and it is not a subtraction either: the market publishes TWO books on the same House question, a binary on control and a distribution of seat bands. They are not the same measure and this house does not convert one into the other, because the band book persistently sums above 100% and therefore serves to discuss the SHAPE of expectations, never to extract a level. What is compared, then, is the BEHAVIOUR of the two in the same week: one stood absolutely still and the other reorganised a quarter of its own mass. That is what the scenarios section measures below.
3.Weighted scenarios for the week
The scenarios below are not narrative: they are read in the prices of contracts the market already publishes. In this issue they use the two control binaries and the distribution of Republican House seat bands. Every number from the distribution comes from the close of 16 October, when the book summed 102.70%, and from the edge of 9 October, when it summed 103.10%, both inside the 95% to 105% gate, and the bands are cited as a SHARE OF THE BOOK, normalised by the sum at that edge. 🔑 The rule that applies here is one of method: a band book summing above 100% carries overpricing, and this house measured across 940 captures that this overpricing is approximately uniform across the bands. Uniform excess means that normalising preserves the internal PROPORTION, and that is why shape can be discussed. It does not mean the normalised share is a calibrated probability, and that is why no number from this distribution is compared with the binary's price. The band of 215 to 219 seats crosses the threshold of 218 and cannot be assigned to either side, so it is the share that stays undefined between the two sides in any reading of shape.
The binary on Democratic control of the House closed the week at 90.50%, exactly where it opened, with floor equal to top across 44 readings. That is the clean reading of the scenario, because the pair sums to 100.00% at both edges and dispenses with any correction. The seats distribution does not give a second price for the same outcome, and it would be a mistake to treat it that way: what it gives is the shape of expectations about HOW MANY seats, a dimension the binary does not touch. And that shape moved a great deal in the week the binary moved not at all. The bands that leave Republicans at 214 seats or fewer went from 85.26% to 90.65% of the book's mass, an advance of 5.39pp of share, and the band of 215 to 219 seats, the only one that crosses the threshold of 218 and therefore the only one that stays undefined between the two sides, shrank from 6.40% to 5.45% of the mass. In words: the seats book concentrated expectations on a more comfortable result for the Democratic side, and at the same time narrowed the zone of doubt about the threshold, without the control price moving a tick.
The binary on Republican control of the House closed at 9.50%, the same value as on 9 October, with floor equal to top and an amplitude of 0.00pp across 44 readings. In the seats distribution, the share of the bands from 220 seats upward fell from 8.34% to 3.89% of the book's mass, meaning it lost more than half its own size in five trading days, and the THREE bands that compose it all gave ground: 220 to 224 seats, 225 to 229, and 230 or more. This is a change of shape and not a second price: it says the seats book came to consider a wide Republican result less likely, and it does not say by how much, because a share of a book carrying overpricing is not a probability. The reorganisation that produced it is concentrated: the band of 205 to 209 seats went from 11.50% to 28.90% raw, an advance of 16.99pp of the book's share that made it the most expensive band in it, with an intraday floor of 11.30% on 11 October at 14:00 UTC and a top of 29.60% on 16 October at 08:00 UTC, an amplitude of 18.30pp across 119 readings. The five bands from 210 seats upward all gave ground, and so did the one below 190 seats, which means the mass came from both tails into the centre.
The contract on the election happening on schedule closed the week at 99.00%, 0.30pp above the 98.70% of 9 October, which leaves the complementary outcome at 1.00%. Within the window it touched 99.10% on 13 October at 18:30 UTC, which is the top of the entire series recorded since 29 July, and on that same 13 October, at 08:30 UTC, it marked the week's floor at 98.20%: the two extremes of the window happened on the same day, an amplitude of 0.90pp across 45 readings. It remains the quietest of the five books this brief follows and the smallest by volume, with USD 0.68M accumulated at the certified reading of 19 October, against USD 10.75M for the most liquid contract in the table.
4.Indicator Grid
| Contract | Current | Δ week | Vol USD acc. | Implied reading |
|---|---|---|---|---|
| House, Democratic control | 90.50% | +0.00pp | USD 10.75M | Floor equal to top at 90.50%, amplitude of 0.00pp across 44 readings, and the same value at the close of all five trading days. The pair summed to 100.00% at both edges: raw and normalised say the same thing. New money in the House book over the period: USD 2.40M, almost four times that of the Senate book. |
| House, Republican control | 9.50% | +0.00pp | USD 7.85M | Also frozen across the 44 readings, with an amplitude of 0.00pp, and with the pair at 100.00% at both edges. Over the same period, the share of the seats book sitting at 220 or more fell from 8.34% to 3.89% of its mass. These are two different things measuring different sides of the same contest, and the brief does not convert one into the other. |
| Senate, Democratic control | 64.50% | +2.00pp raw, +0.73pp normalised | USD 4.06M | It rose in two steps, 63.50% on 11 October and 64.50% on 15 October at 23:30 UTC, held through Friday. The pair sum went from 99.00% to 101.00% over the period, and that is why the two measures disagree by 1.27pp. In the previous issue the same contract had the raw figure overstating a FALL for the opposite reason, with the sum going from 100.00% to 99.00%. |
| Senate, Republican control | 36.50% | +0.00pp raw, -0.73pp normalised | USD 3.09M | The week's cleanest reading on why price is not probability: the contract closed where it opened, 0.00pp across 49 readings, and its implied probability gave back 0.73pp. Intraday top of 37.50% on 10 October at 03:30 UTC and floor of 35.50% on 9 October at 03:00 UTC. |
| Election happens on schedule | 99.00% | +0.30pp | USD 0.68M | Top of the entire series at 99.10%, touched on 13 October at 18:30 UTC, and the week's floor at 98.20% on the same day at 08:30 UTC. The path is not monotonic: 98.50% on 10 October, 98.40% on 11, and then three days rising to 99.10% on 14 October, repeated at the close of 15. |
| Republican House seats, distribution of 10 bands | sum 102.70% at the close | from 103.10% on 9 October | USD 0.33M | Inside the 95% to 105% gate at both edges of the week, with a clean partition across the ten bands, no overlap and no gap. It is the lowest volume book in this table, with USD 0.01M of new money in the week, and at the same time the one that reorganised the most: the band of 205 to 209 seats advanced 16.99pp of the book's share and became the most expensive band in it. |
5.Liquidity and market structure
6.Electoral calendar through November
| Date | Sample | Why it matters | |
|---|---|---|---|
| 3/Nov, Tuesday | Midterm election day | 435 House seats and one third of the Senate | It is the only date on this grid that settles the contracts. Every price cited in this issue is the implied probability of an outcome that becomes known on that day. |
| 12/Oct, Monday | Musk becomes the cycle's largest known donor, with USD 276 million declared | declared donations for the cycle | On that same 12 October the band of 205 to 209 Republican seats jumped from 11.30% to 20.00% raw, its largest daily advance of the week. A same day coincidence of dates does not establish direction, and the series does not allow saying which reading came first within the day, so the brief records both and attributes nothing. |
| 13/Oct, Tuesday | Trump flies 3,400 miles to Alaska to try to save a Senate seat | 1 state Senate contest | The contract on Democratic control of the Senate had risen 1.00pp on 11 October, TWO days before this publication, and stood still at 63.50% on the 13th and 14th. The order of events does not support a causal relationship in either direction. |
| 15/Oct, Thursday | Reports on the president's attacks on voting and on suspicious mailings sent to election officials | 2 outlets on the same day | The contract on the election happening on schedule CLOSED that day at 99.10%, which is the highest close of the week, tied with that of 14 October, and the top of the entire series had been touched two days earlier. That is, on the day of the two reports the book that would price calendar risk was at its highest point. The brief records the coincidence and does not interpret it. |
| 14/Oct, Wednesday | Election advertising spending passes USD 10 billion and keeps rising | national advertising market | This is context on the volume of resources in the contest, and not a price event. On the same day the band of 205 to 209 seats retreated from 21.90% to 16.80% raw, its only retreat of the week, before rising again over the following two trading days. |
The grid above has one date that settles a contract, 3 November, and four that settle none. The four are included because they are verifiable and dated, and they come with the measurement alongside when one exists. In three of them the series does not support the link between the fact and the price, and in those cases the brief says so and stops there. The 15 October case is the week's most instructive precisely because the price moved in the opposite direction to what a hurried reading would expect.
7.Watch list, week triggers
- The band of 205 to 209 seats, at 28.14% of the book's mass and now the most expensive in it It was 11.15% of the book's mass at the edge of 9 October and came to concentrate more than a quarter of the entire seats book. As long as it absorbs mass from the bands below 200, the share of the book sitting below the majority threshold rises without the binary needing to move, which is exactly what happened this week. If it gives mass back to the band of 215 to 219, the shape loses definition again precisely where the question is decided.
- The House pair sum, at exactly 100.00% at both edges of the week and at 101.00% at Monday's certified reading Across the five trading days the House was the book in which raw and normalised coincided, and that is why its delta did without the second measure. At the certified reading of 19 October at 18:33 UTC the Democratic side stands at 91.50% and the Republican at 9.50%, summing 101.00%. If that sum persists, the House delta will need both measures, as the Senate's already does. And the Senate did the inverse over the same interval: it closed the week summing 101.00% and stands at 100.00% at that same certified reading.
- The band of 215 to 219 seats, at 5.45% of the book's mass It is the only band in the book that crosses the threshold of 218, and therefore the only share that cannot be assigned to either side: the larger it is, the more undefined the book's shape is near the threshold. It fell from 6.40% to 5.45% of the mass over the week, which means the book reduced its own zone of doubt about who takes the majority, at the same time as it shifted mass to well below the threshold. If it gets more expensive again, the shape loses resolution exactly at the point that decides the question.
- The dispersion of 10.00pp between pollsters in the generic ballot window Within the same 30 day window, Quinnipiac measured a Democratic advantage of 12.00 points with field closing on 27 September and 1,032 registered voters, and Harvard/Harris measured 2.00 points with field closing on 28 September among likely voters. That is ten points between two first tier pollsters with practically simultaneous field dates, and it is a number the average on its own hides. The three rounds with field dates inside this issue's own week are tighter among themselves: 9.00 points from Beacon and Shaw on 12 October, 8.00 from RMG Research and 6.00 from ActiVote, the latter two on 14 October, a range of 3.00 points.
8.Methodology
Prices read on the AFOS's own route with the country parameter and an uncached reading, and confirmed by the capture guard: two readings 8 minutes apart, released only if they agree within 0.20pp, with the two required timestamps distinct. This issue's certified reading is from 19 October 2026 at 18:33:27 UTC, and the real interval between the two timestamps was 8.0 minutes. Each contract's series is read in the database backup and never on the history route, because that route caps the window at 90 days and in the Senate contract that would make the series' real floor, 39.50%, appear as a higher value. The stretch of series the daily backup does not yet cover is closed with the route's intraday points and checked by positive control, series by series, comparing the two sources over the window in which both have data: in this issue the five series closed with identical points on both sides, and that is why every superlative here holds for the entire series and not only for what the backup already had. Closes are the last point recorded on each day, and top and floor are intraday, because a daily close hides extremes. The week's delta is anchored to the previous Friday's close, 9 October, which is the convention of this United States series, and the tool that computes it was required to reproduce the six numbers published in the previous issue before being used here. Any band distribution is only cited when the book's sum falls between 95% and 105% at the edges used, and it is cited as a share of the book, normalised by the sum at that edge. A band book summing above 100% carries overpricing, and this house tests where it sits by comparing the normalised share of the centre against the raw sum along the series: if the centre's share falls when the sum rises, the excess is in the tail and normalising uniformly does not remove it; if it rises with it, the excess is in the centre; with no clear relationship, it is approximately uniform. Measured across 940 captures of this book, the correlation is 0.267, that is, the excess is approximately uniform, and in that case normalising preserves the internal proportion. That is why the distribution here supports a statement about SHAPE and never about level, and that is why none of its shares is compared with the price of a binary contract. The generic ballot average is a simple arithmetic one, unweighted, in a 30 day rolling window, with one round per pollster per wave by the hierarchy likely voter, registered voter, adults, and with a wave published under another instrument kept out of the count and declared with the pollster's own wording. Book depth is not cited in this brief by an editorial decision of May 2026.
9.Additional reading · macro coverage
The items below are the outlets cited in the sections above, with the address of the article itself. Items the collection reached only through the aggregator were discarded from this list, because the aggregator's address does not point to the outlet.
- The New York Times, 12/Oct · The breakdown of Musk's USD 276 million in the attempt to preserve the Republican majority, www.nytimes.com/2026/10/12/us/politics/elon-musk-midterms.html
- The Guardian, 13/Oct · The same donor, already at USD 366 million in the cycle, by the outlet's count, www.theguardian.com/us-news/2026/oct/13/elon-musk-political-donor-midterm-elections
- The Washington Post, 12/Oct · The bellwether races to watch in November, www.washingtonpost.com/politics/2026/10/12/bellwether-races-watch-novembers-midterm-elections/
- The Washington Post, 13/Oct · Trump flies to Alaska to try to save a Senate seat, www.washingtonpost.com/politics/2026/10/13/trump-fly-3400-miles-alaska-hopes-saving-senate-seat/
- The New York Times, 14/Oct · Election advertising spending passes USD 10 billion, www.nytimes.com/2026/10/14/us/politics/midterms-ad-spending-10-billion.html
- The New York Times, 15/Oct · The president's attacks on voting and fears of a challenge to the result, www.nytimes.com/2026/10/15/us/politics/trumps-attacks-voting-midterms.html
- NPR, 15/Oct · Suspicious mailings sent to election officials and to congressional offices, www.npr.org/2026/10/15/nx-s1-6003857/suspicious-mailings-election-congressional-offices
- The Hill, 13/Oct · Democrats ahead on favourability and on the economy, by the Gallup measurement, thehill.com/homenews/campaign/6145986-democrats-lead-republicans-on-favorability-economy-gallup/
- The Hill, 15/Oct · Brit Hume says a broad Republican defeat is possible in the election, thehill.com/homenews/campaign/6151939-gop-midterm-washout-risk/
- The New York Times, 16/Oct · A transfer of USD 1 million to the Michigan Senate campaign, www.nytimes.com/2026/10/16/us/politics/aoc-el-sayed-million-transfer-midterms.html
- NewsNation, 15/Oct · The cycle as the most expensive on record, by the AdImpact estimate, www.newsnationnow.com/politics/2026-midterm-elections/2026-midterm-most-expensive-ever-adimpact/
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.