AFOS Weekly · USA-2026 midterms Political Risk Weekly

AFOS Analytics

Real-time political pricing
Prediction market × polls × press
no smoothed averages

Harvard Dataverse · DOI 10.7910/DVN/XRUT8U
United States·Issue №1·Week of 2026-07-30 to 2026-08-06·published thursday 07:00 BRT

What the prediction market did

The eight midterm contracts here hold USD 17.01M, and one price did nearly all the work. Democrats to control the House went to 87.50%, two points above its first six days. Both Senate prices stood still today, Republicans at 54.50% and Democrats at 44.50%. The one other published price, on whether the election happens on schedule, hit its lowest print yet.

House · Democrats

87.50%

↑ 1.00pp on Aug 6

Deepest book, USD 5.01M. It printed 85.50% daily from July 29 to August 3, went to 86.50% on August 4, then stepped again today.

House · Republicans

13.50%

→ 0.00pp, all 13 readings

On USD 4.05M. The quiet finding: this side printed the same number at every reading since the series opened, while the Democratic side moved twice.

Senate · Republicans

54.50%

→ 0.00pp, second day

On USD 1.53M, after coming off 55.50% on August 5. Over 30 days recorded the range is 52.50% to 56.50%.

Senate · Democrats

44.50%

→ 0.00pp since Aug 2

On USD 2.21M. It rose from 43.50% on August 2 and held, so this book's two sides now add to 99.00%.

The fifth published price, on whether the election happens on schedule, sits at 95.65%, down 0.55pp and the lowest print of its series, on USD 0.31M, the smallest book of the set.

What the polls did

The disagreement comes before the average, because it is the finding. Twenty-four institutes measured the same race in the same window and came back 9 points apart. The new fieldwork narrowed nothing: the average of 31 polls is D+5.55, and the newest, Quantus Insights through August 4, reads D+6.

Widest

Big Data Poll · D+11

Narrowest

Reuters/Ipsos · D+2

Spread between institutes

9 points

Big Data Poll polled 2,617 likely voters through July 29. Reuters/Ipsos polled 1,003 registered voters through July 26. Neither is small and neither is wrong: houses assume different things about who shows up, and that assumption is most of the gap. Median D+5. Reuters/Ipsos then returned with 3,542 respondents through August 3 at D+5, three points above its own reading a week earlier.

What the press reported

What the fixed-list outlets covered, July 30 to August 6

The week had an actual event. On August 4 Michigan voted, and Abdul El-Sayed beat Stevens in the Democratic Senate primary. AP News called it a narrow victory over the Democratic establishment on August 5. The Washington Post was blunter the same day: the establishment has lost its grip. The New York Times called it the left's biggest test of the fall.

Underneath it ran a slower story with nothing to do with polling. NBC News reported on July 30 that Democratic Senate candidates in rural states are dropping out, clearing the field for independents. Fox News reported one such exit on August 5 in South Dakota, leaving an independent as the lone challenger to Senator Mike Rounds.

📌 Worth noticing: a real result landed in a watched race, and the two contracts deciding control of Congress moved one point and zero. A primary is one seat in one state, and the national picture is a different measurement.

Our measurement: That is CNN's own poll, fielded through July 27, and our file carries it at D+8. Our average across the 31 polls in the window is D+5.55: three sit above D+8, twenty-four below. The headline is accurate about its own poll, and it is one point inside a nine-point spread.

The crossing of the week

The two measurements moved in opposite directions this week

The market grew more confident about Democrats in the House, from 85.50% to 87.50% in three days. The poll average moved the other way, from D+5.75 to D+5.55, and that fall came from real new fieldwork, not from old polls ageing out of the window. Both are small, and they point opposite ways across the same seven days. This brief will not subtract one from the other: they measure different things in different units.

The comparison we ran last week cannot be run this week

Two Senate contracts answer the same question by different routes: one on control of the chamber, one on how many seats Republicans win. Last week they nearly agreed, and this brief said so. This week the seat-count bands add to 105.75%, and probabilities that sum past 100 are not probabilities. A band market reaches this page only when its bands total 95% to 105%, so the comparison is off. Saying so beats publishing arithmetic that does not close. The House seat-count market fails worse, at 109.50%.

How to read this number

Why this brief never subtracts the prediction market from the polls

The market prices the probability that a party controls a chamber. A poll measures a lead in vote points. Different quantities, so 87.50% minus 5.55 is a number with no unit, and it would change size if the poll used another scale. The American reason it matters: in 2012 Democrats won more votes and still won fewer seats. This brief cross-reads direction and movement, never level.

Sources

Prediction market

Polling

Press, outlets cited in sections 4 and 5

Method

Prediction market: Polymarket, read through this platform's own proxy and captured every 30 minutes, with a guaranteed daily point even when the price does not move, so the series records time and not just changes. Every price quoted here survived a capture guard: two independent readings eight minutes apart, released only if they agree within 0.20pp. The House series opened on July 29, so a high here means the highest of nine days, never of the cycle. Polls: a simple unweighted average over a 30-day window, currently 31 polls from 24 institutes with field dates from July 7 to August 4, no institute excluded, built from an index that says which polls exist while every number stays attributed to the institute that produced it. Poll aggregators are deliberately left out, because averaging an aggregator would mean averaging someone else's model. When an institute publishes several cuts of one round, one enters, by likely voters before registered voters before adults. Press: a fixed list of 22 outlets, chosen once and applied identically on every run, so no outlet is picked after the news breaks. Fifteen are read from their own RSS, which gives the outlet's canonical article link; the rest come through an aggregator, and those links are shown as headlines rather than addresses because the address would not be the outlet's own. AFOS does not classify outlets by political leaning. Distributions: a band market only reaches this page when its bands add to between 95% and 105%. Today the governors market passes at 101.00% and the turnout market at 100.70%. The Senate seat-count market fails at 105.75% and the House seat-count market at 109.50%, so neither appears as probability. The popular-vote margin market adds to 152.05% and is structurally incapable of closing, because one of its bands is an 'any other outcome' at 45.50% while the rest are cumulative; it is collected daily anyway, because it is the one contract that would allow a direct comparison with the polls if it were ever a partition. What this brief does not do: it does not forecast, it does not say which measurement is right, and it does not tell anyone how to vote.

AFOS
AFOS Analytics
Global Political Risk Intelligence · GLOBAL BY DESIGN
www.afos-analytics.com · contact@afos-analytics.com · github.com/AFOS-Analytics

Get it by email

AFOS Weekly follows the US midterms every Thursday, written for voters: where the prediction market, the polls and the press disagree, and what each one measures.

No spam. Unsubscribe anytime.

After signing up you choose whether to receive it in English, Portuguese or Spanish.

AFOS Weekly · Issue No. 1 · The market grew more certain this week. The first new polls in a week went the other way. | AFOS Analytics