Executive summary · four days to Election Day

Democrats take the House. The Senate ties 50–50, and Maine decides who runs it.

We call the 27 most competitive races: all 17 House toss-ups and the 10 Senate races Cook rates toss-up or lean. A Democratic national environment of D+10.0 is applied to each seat's 2024 result, adjusted for measured candidate strength. Digital Atlas adds what the arithmetic cannot: what each district is made of, what its economy is doing and what voters there are dealing with.

House218 for control

Democratic majority

225 D·202 R·8 unresolved

218 for control

220

198

Rated Lean or saferOur toss-up callNo model call

Democrats reach 220 seats on races Cook rates Lean D or safer, before a single toss-up is counted.
Of the 17 toss-ups, our model calls 9: 5 D, 4 R. The other 8 were redrawn for 2026 and have no comparable 2024 result, so we leave them unresolved.
Where a public poll and the Crystal Ball rating agree, we add an external lean: CA-22, NC-01, TX-15 lean D, giving 228 D / 202 R with 5 coin flips.

Senate51 for control, or 50 + Vice President

50–50 tie: Republicans keep control on the Vice President's vote

50 D·50 R·net D +3

50 · VP breaks ties

Not up in 2026Non-target, by ratingOur target call

Democrats gain 3: AK and OH-special on our model, NC on its Cook rating. They hold their open seats in MI, MN, NH; Republicans hold IA, KS, ME, TX, SC.
Maine is the tipping seat. Our model has Collins ahead by 0.7 points; at the raw generic-ballot shift (D+11.3) she trails by 0.6 and Democrats win 51–49.
The risk leans Democratic: public polls lead D in KS (D+10.0) and ME (D+4.0), two seats our model gives to Republicans.

House toss-ups175 D · 4 R · 8 no model call

Senate targets105 D · 5 R · 7 toss-ups + 3 lean

National environmentD+10.0Generic ballot D+8.7 vs 2024 House R+2.6; Trump approval 38.6%

Confidence across 19 calls11 Mod · 8 LowNo High calls: no dated market prices were captured, and High requires market agreement

Our House calls model margin · poll

PA-08Cognetti v BresnahanD flipD+8.4ModerateD+5.0
MI-07Lawrence v BarrettD flipD+5.0ModerateD+8.0
MI-10Hines v BouchardD openD+3.9ModerateR+4.8
MT-01Forstag v FlintD openD+2.3ModerateR+2.0
NY-17Conley v LawlerD flipD+0.2LowD+2.0
VA-01Taylor v WittmanR holdR+2.8ModerateD+4.0
MI-04McCann v HuizengaR holdR+4.8ModerateR+3.2
IA-02James v MitchellR openR+5.6LowD+3.0
KY-06Dembo v AlvaradoR openR+16.1ModerateD+6.0
CA-22Villegas v ValadaoNo model call—Ext. lean DD+11.0
TX-15Pulido v De La CruzNo model call—Ext. lean DD+7.0
NC-01Davis v BuckhoutNo model call—Ext. lean DD+1.8
OH-09Kaptur v MerrinNo model call—Coin flipD+6.0
OH-07Poindexter v MillerNo model call—Coin flipD+4.0
FL-14Castor v BeltranNo model call—Coin flipD+0.9
FL-22Dandiya v AskarNo model call—Coin flipD+0.8
TX-35Garcia v De La CruzNo model call—Coin flipR+1.0

Our Senate calls model margin · poll

MNFlanagan v TafoyaD openD+14.2ModerateD+14.0
NHPappas v SununuD openD+12.8ModerateD+7.0
AKPeltola v SullivanD flipD+12.0LowD+2.0
OH-specialBrown v HustedD flipD+9.0LowD+3.0
MIEl-Sayed v RogersD openD+8.6ModerateD+7.0
MEJackson v CollinsR holdR+0.7LowD+4.0
IATurek v HinsonR openR+3.2LowR+2.0
TXTalarico v PaxtonR openR+3.7LowR+1.0
KSHamilton v MarshallR holdR+6.1LowD+10.0
SCAndrews v GrahamR holdR+7.9ModerateEven

Margins are Democratic minus Republican vote share. “Ext. lean” is analyst judgment for redrawn districts where the model cannot run; it is not counted in the model totals. Polls are shown for comparison and never enter a margin.

Where we could be wrong

Maine sits inside the toss zone. R+0.7 in our model; the CNN poll has Jackson ahead by 4. A national environment of D+10.7 or more flips it, and with it the Senate.

Polls run more Democratic than the model. In 5 of 9 called House races the latest poll points the other way (IA-02, KY-06, MI-10, MT-01, VA-01). Two of the model's Republican calls, IA-02 and KY-06, are open seats whose baseline still carries the departing Republican's 2024 vote.

Two Democratic Senate gains rest on candidate brand. Without the measured personal votes of Peltola (+15.1) and Brown (+10.2), AK and OH-special would be Republican. Their polls agree on direction but are tighter (AK D+2.0, OH-special D+3.0).

What Digital Atlas added

10.3Mplaces profiled across 50 states

466,754places mapped to the exact 2026 lines of 21 districts

236,874jobs traced block by block in AZ-06

18state labor series with dated source receipts

Atlas tells us why a seat is competitive and what its voters face: costs, health care and, in 11 of 27 races, data centers and power bills. It never adds points to a margin. That separation is deliberate, so the call stays auditable and place data is never mistaken for voter preference.

See the Atlas context →

Part 1 · Context

The battlefield

A midterm is a referendum on the president. All 435 House seats and 35 Senate seats are on the ballot, the president is not, and turnout falls to roughly 35–50% of eligible voters, older and more partisan than in a presidential year. Since 2002 the president's party has lost House seats in five of six midterms.

The national environment, as of 2026-10-28

Generic ballotD+8.7Silver Bulletin average · 2024 House vote was R+2.6

Trump approval38.6%Every wave midterm since 2002 came with approval at 45% or below

Shift applied to every raceD+10.0Chosen judgment parameter; the raw generic-ballot shift is D+11.3

CPI inflation, Sep 20263.72%

Unemployment, Sep 20264.2%

Shelter inflation, Sep 20263.22%

Inflation, unemployment and shelter are computed from raw FRED series (CPIAUCSL, UNRATE, CUSR0000SAH1), January 2023 to September 2026.

Six midterms of evidence

Cycle	President · approval	House	Senate	The close races that decided it
2002	G.W. Bush (R) · ~63%	+8 R	+2 R	Post-9/11 rally. MN: Norm Coleman (R) won after Paul Wellstone’s death; MO: Jim Talent (R) +1.3; NH: John Sununu (R) +3.9. One of only two modern midterms where the president’s party gained.
2006	G.W. Bush (R) · ~38%	−30 R	−6 R	Iraq + Katrina. Senate flips: MT Tester (D) +0.9, VA Webb (D) +0.4, MO McCaskill (D) +2.4, RI Whitehouse (D) +7, OH Brown (D) +12, PA Casey (D) +17. House: 22 Republican incumbents lost; no Democrat incumbent lost.
2010	Obama (D) · ~45%	−63 D	−6 D	Worst jobs crisis in decades. Senate flips: PA Toomey (R) +2.0, WI Johnson (R) +4.9, IL Kirk (R) +1.6, AR Boozman (R) +21, ND Hoeven (R), IN Coats (R). Held by D: NV Reid +5.6, CO Bennet +1.7, WA Murray +4.9. Largest House swing since 1948.
2014	Obama (D) · ~42%	−13 D	−9 D	Approval-driven second-term slump despite improving economy. Senate flips: AR Cotton (R) +17, AK Sullivan (R) +2.8, CO Gardner (R) +2.5, IA Ernst (R) +8.5, LA Cassidy (R) +13 runoff, MT Daines (R), NC Tillis (R) +1.5, SD Rounds (R), WV Capito (R) +28.
2018	Trump (R) · ~41%	−40 R	+2 R	Strong economy did not prevent a House wave. Senate: D lost ND (Cramer +10.8), MO (Hawley +5.8), IN (Braun +5.9), FL (Scott +0.1); D gained NV (Rosen +5.0), AZ (Sinema +2.4). Held: TX Cruz (R) +2.6, MT Tester (D) +3.5, WV Manchin (D) +3.3.
2022	Biden (D) · ~41%	−9 D	+1 D	Highest inflation in 40 years, but Dobbs + candidate quality limited losses. Senate: PA Fetterman (D) +5.0 flip, GA Warnock (D) +2.8 runoff, NV Cortez Masto (D) +0.8, AZ Kelly (D) +5.0. Held by R: WI Johnson +1.0, NC Budd +3.4, OH Vance +6.2, FL Rubio +16.6.

Approval sets the size of the swing more reliably than the economy: 2018 had 3.8% unemployment and still cost Republicans 40 House seats. Six cycles support pattern-matching, not a fitted model.

House: 435 districts, 17 toss-ups

Solid D 186Likely D 18Lean D 16Toss Up 17Lean R 17Likely R 13Solid R 168

2026 district lines (Census TIGER CD120) coloured by Cook ratings of October 23. Toss-ups are the light districts; hover for candidates and our call. IL-11 is missing from the display geometry and DC-98 is a separate non-voting district; all 17 toss-ups are drawn.

Senate: 35 races, 10 targets

Our call DOur call RD, low confidenceR, low confidenceNot a target, rated DNot a target, rated RNo race in 2026

Today the Senate is 53 R, 47 D. Democrats need +3 for a 50–50 tie, which the Republican Vice President breaks, and +4 for a majority. Both Class III specials (Ohio, Florida) are counted once.

The map expanded. Five Senate targets (Ohio, Texas, Iowa, Alaska and Kansas) are states Trump won by double digits in 2024. These are the races where a uniform national swing is least reliable, so the rules hold them at Low confidence unless a dated market agrees.

Who is running

Race	Candidates	Seat	Cook	Crystal Ball	Local issues in coverage
PA-08	Paige Cognetti (D) vs Rob Bresnahan Jr (R, inc.)	Incumbent Rob Bresnahan Jr (R)	Toss Up	Lean D	costs/healthcare/ACA/Medicaid; data centers/AI; corruption/stock trading
MI-07	William Lawrence (D) vs Tom Barrett (R, inc.)	Incumbent Tom Barrett (R)	Toss Up	Toss-up	Mason data center/local control/grid costs; prices/tariffs; healthcare
MI-10	Christina Hines (D) vs Michael Bouchard (R)	Open seat	Toss Up	Toss-up	healthcare/money in politics; tariffs/manufacturing; border/crime
MT-01	Sam Forstag (D) vs Aaron Flint (R)	Open seat	Toss Up	Toss-up	housing/healthcare/fuel; public lands/timber; data centers
NY-17	Cait Conley (D) vs Mike Lawler (R, inc.)	Incumbent Mike Lawler (R)	Toss Up	Toss-up	economy/housing; immigration/sanctuary policy; Israel/antisemitism/Iran
VA-01	Shannon Taylor (D) vs Rob Wittman (R, inc.)	Incumbent Rob Wittman (R)	Toss Up	Toss-up	affordability/groceries/housing; data centers/electricity; healthcare/ACA/rural hospitals
MI-04	Sean McCann (D) vs Bill Huizenga (R, inc.)	Incumbent Bill Huizenga (R)	Toss Up	Toss-up	energy affordability/data centers; gas/cost of living
IA-02	Lindsay James (D) vs Joe Mitchell (R)	Open seat	Toss Up	Lean D	cost of living/gas/housing/childcare/healthcare; taxes/regulation/tariffs
KY-06	Zach Dembo (D) vs Ralph Alvarado (R)	Open seat	Toss Up	Toss-up	affordability/tariffs/housing; Medicaid/rural hospitals; data centers/utility costs
CA-22	Randy Villegas (D) vs David Valadao (R, inc.)	Incumbent David Valadao (R)	Toss Up	Lean D	Medicaid/health care; water/agriculture; affordability/gas
TX-15	Bobby Pulido (D) vs Monica De La Cruz (R, inc.)	Incumbent Monica De La Cruz (R)	Toss Up	Lean D	costs/healthcare/economic opportunity; immigration/labor; water/1944 treaty
NC-01	Don Davis (D, inc.) vs Laurie Buckhout (R)	Incumbent Don Davis (D)	Toss Up	Lean D	costs/jobs/rural healthcare; NC12/flood insurance/coastal erosion; disaster recovery/fishing
OH-09	Marcy Kaptur (D, inc.) vs Derek Merrin (R)	Incumbent Marcy Kaptur (D)	Toss Up	Toss-up	costs/taxes/tariffs/energy; data centers
OH-07	Brian Poindexter (D) vs Max Miller (R, inc.)	Incumbent Max Miller (R)	Toss Up	Toss-up	costs/fuel/tariffs/healthcare; workers' bargaining rights; ethics allegations/investigation
FL-14	Kathy Castor (D, inc.) vs Mike Beltran (R)	Incumbent Kathy Castor (D)	Toss Up	Toss-up	property/flood insurance; housing/utility/health costs; abortion/immigration
FL-22	Pia Dandiya (D) vs Casey Askar (R)	Open seat	Toss Up	Toss-up	affordability/housing; ACA/health care; education/tariffs/data centers
TX-35	Johnny Garcia (D) vs Carlos De La Cruz (R)	Open seat	Toss Up	Toss-up	grocery/healthcare premiums; immigration/tariffs/employers; San Antonio military installations
MN	Peggy Flanagan (D) vs Michele Tafoya (R)	Open; Tina Smith retiring.	Lean D	Lean D	program-fraud oversight; taxes; immigration
NH	Chris Pappas (D) vs John E. Sununu (R)	Open; Jeanne Shaheen retiring.	Lean D	Likely D	cost of living; tariffs; Iran war
AK	Mary Peltola (D) vs Dan Sullivan (R, inc.)	Incumbent Dan Sullivan (R)	Toss Up	Lean D	affordability/energy; fuel costs; fisheries/bycatch
OH-special	Sherrod Brown (D) vs Jon Husted (R, inc.)	Class III special; appointed incumbent Jon Husted; remainder of the Vance term through January 2029.	Toss Up	Toss Up	affordability/prescriptions/utility costs; data-center incentives; immigration
MI	Abdul El-Sayed (D) vs Mike Rogers (R)	Open; Gary Peters retiring.	Toss Up	Lean D	affordability; healthcare; Iran war
ME	Troy Jackson (D) vs Susan Collins (R, inc.)	Incumbent Susan Collins (R)	Toss Up	Toss Up	cost of living/Medicaid; heating energy
IA	Josh Turek (D) vs Ashley Hinson (R)	Open; Joni Ernst retiring.	Toss Up	Toss Up	tariffs; energy; healthcare costs
TX	James Talarico (D) vs Ken Paxton (R)	Open November ballot, R-held after Cornyn lost the runoff.	Toss Up	Toss Up	gas/grocery/housing; utility/data centers; healthcare
KS	Adam Hamilton (D) vs Roger Marshall (R, inc.)	Incumbent Roger Marshall (R)	Toss Up	Toss Up	tariffs; Iran/fuel prices; ACA/healthcare
SC	Annie Andrews (D) vs Darline Graham (R, inc.)	Regular Class II race; appointed incumbent Darline Graham; not an open seat.	Lean R	Lean R	Social Security; data centers; healthcare

Ratings: Cook Political Report via dated 270toWin mirrors (House Oct 23, Senate Oct 20); Crystal Ball dated Oct 27–28, comparison only. Local issues come from the consolidated race evidence notes.

Part 3 · Congress

Race calls

Every margin here comes from one generated file, race_results.json, built by the shared model code. Nothing below is hand-entered. Polls, ratings and markets sit beside the model for comparison and never move a number.

House accounting

Layer	D	R	Open
Cook Lean or safer (Oct 23)	220	198	17
+ model calls on 9 toss-ups	+5	+4	−9
Model allocation	225	202	8
+ external leans (CA-22, NC-01, TX-15)	+3	+0	−3
With external leans	228	202	5

Coin flips: FL-14, FL-22, OH-07, OH-09, TX-35. Today the House is 218 R, 214 D with vacancies.

Senate accounting

Component	D	R
Not up in 2026	34	31
25 non-target races, by Cook rating	11	14
10 target races, by our model	5	5
Total	50	50
At the raw D+11.3 shift	51	49

Seat changes: Democrats gain AK, NC, OH-special; Republicans gain none. A 50–50 Senate is run by the party of the Vice President.

All 27 races: model against the latest poll

Model call DModel call RRaw-shift alternative (D+11.3)Latest public poll±5.7 pts, the backtest's median miss±2 toss zone

Hover a row for the arithmetic and the poll's sponsor and dates. The ±5.7 band is the median absolute error from the 2018 and 2022 Senate backtest; it is a scale for judgment, not a confidence interval.

House toss-ups

Race	2024 baseline	Candidate	Model margin	Call	Latest poll (comparison)	Ratings	Watch
PA-08 Paige Cognetti (D) vs Rob Bresnahan Jr (R, inc.)	R+1.6 2024 House	+0.0 no measurement	D+8.4 alt D+9.7	D flip Moderate	D+5.0 New York Times / Philadelphia Inquirer / Siena 2026-10-16–2026-10-22	Cook Toss Up CB Lean D	—
MI-07 William Lawrence (D) vs Tom Barrett (R, inc.)	R+3.7 2024 House	-1.3 one race, halved	D+5.0 alt D+6.3	D flip Moderate	D+8.0 Emerson / Poll Vault 2026-10-12–2026-10-14	Cook Toss Up CB Toss-up	—
MI-10 Christina Hines (D) vs Michael Bouchard (R)	R+6.1 2024 House	+0.0 open seat	D+3.9 alt D+5.2	D open Moderate	R+4.8 RedBox Research / A Fight Worth Having super PAC 2026-10-20–2026-10-23	Cook Toss Up CB Toss-up	poll disagreesopen seat: R winner's 2024 vote in baseline
MT-01 Sam Forstag (D) vs Aaron Flint (R)	R+7.7 2024 House	+0.0 open seat	D+2.3 alt D+3.6	D open Moderate	R+2.0 RedBox Research / A Fight Worth Having super PAC 2026-10-20–2026-10-23	Cook Toss Up CB Toss-up	poll disagreesopen seat: R winner's 2024 vote in baseline
NY-17 Cait Conley (D) vs Mike Lawler (R, inc.)	R+6.3 2024 House	-3.5 one race, halved	D+0.2 alt D+1.5	D flip Low	D+2.0 Emerson / PIX11 2026-09-27–2026-09-29	Cook Toss Up CB Toss-up	inside toss zone
VA-01 Shannon Taylor (D) vs Rob Wittman (R, inc.)	R+12.8 2024 House	+0.0 no measurement	R+2.8 alt R+1.5	R hold Moderate	D+4.0 Washington Post-Schar School 2026-09-23–2026-09-29	Cook Toss Up CB Toss-up	poll disagrees
MI-04 Sean McCann (D) vs Bill Huizenga (R, inc.)	R+11.7 2024 House	-3.1 one race, halved	R+4.8 alt R+3.5	R hold Moderate	R+3.2 RedBox Research / A Fight Worth Having super PAC 2026-10-20–2026-10-23	Cook Toss Up CB Toss-up	—
IA-02 Lindsay James (D) vs Joe Mitchell (R)	R+15.6 2024 House	+0.0 open seat	R+5.6 alt R+4.3	R open Low	D+3.0 Emerson / Poll Vault 2026-10-12–2026-10-14	Cook Toss Up CB Lean D	poll disagreesopen seat: R winner's 2024 vote in baseline
KY-06 Zach Dembo (D) vs Ralph Alvarado (R)	R+26.1 2024 House	+0.0 open seat	R+16.1 alt R+14.8	R open Moderate	D+6.0 Public Policy Polling for House Majority PAC super PAC 2026-10-19–2026-10-20	Cook Toss Up CB Toss-up	poll disagreesopen seat: R winner's 2024 vote in baseline
CA-22 Randy Villegas (D) vs David Valadao (R, inc.)	none No comparable CD120 House baseline: official precinct/block recast is incomplete, including masked small-precinct votes and Madera issues; no authoritative House recast was found	—	—	No model call Ext. lean D	D+11.0 USC California battleground poll 2026-10-22–2026-10-28	Cook Toss Up CB Lean D	—
TX-15 Bobby Pulido (D) vs Monica De La Cruz (R, inc.)	none No comparable CD120 House baseline: the official PLANC2333 recast covers president, Senate and other statewide contests, not U.S. House; statewide races are not substituted	—	—	No model call Ext. lean D	D+7.0 Texas Southern for Express-News / El Pais / Puente 2026-10-13–2026-10-16	Cook Toss Up CB Lean D	—
NC-01 Don Davis (D, inc.) vs Laurie Buckhout (R)	none No comparable CD120 House baseline: added territory voted in old NC-03 with no Democrat, so party-vote aggregation is not a defensible contested-House baseline	—	—	No model call Ext. lean D	D+1.8 RedBox Research / A Fight Worth Having super PAC 2026-10-20–2026-10-23	Cook Toss Up CB Lean D	—
OH-09 Marcy Kaptur (D, inc.) vs Derek Merrin (R)	none No comparable CD120 House baseline: split-county recasting is required and no official House-only precinct recast was found	—	—	No model call Coin flip	D+6.0 RedBox Research / A Fight Worth Having super PAC 2026-10-20–2026-10-23	Cook Toss Up CB Toss-up	—
OH-07 Brian Poindexter (D) vs Max Miller (R, inc.)	none No comparable CD120 House baseline: split-county recasting is required and no official House-only precinct recast was found	—	—	No model call Coin flip	D+4.0 RedBox Research / A Fight Worth Having super PAC 2026-10-20–2026-10-23	Cook Toss Up CB Toss-up	—
FL-14 Kathy Castor (D, inc.) vs Mike Beltran (R)	none No comparable CD120 House baseline: official boundary packets do not supply a House-only 2024 recast	—	—	No model call Coin flip	D+0.9 RedBox Research / A Fight Worth Having super PAC 2026-10-20–2026-10-23	Cook Toss Up CB Toss-up	—
FL-22 Pia Dandiya (D) vs Casey Askar (R)	none No comparable CD120 House baseline: official boundary packets do not supply a House-only 2024 recast	—	—	No model call Coin flip	D+0.8 RedBox Research / A Fight Worth Having super PAC 2026-10-20–2026-10-23	Cook Toss Up CB Toss-up	—
TX-35 Johnny Garcia (D) vs Carlos De La Cruz (R)	none No comparable CD120 House baseline: the official PLANC2333 recast covers president, Senate and other statewide contests, not U.S. House; statewide races are not substituted	—	—	No model call Coin flip	R+1.0 Braun for Washington Post / George Mason 2026-10-14–2026-10-19	Cook Toss Up CB Toss-up	—

The 8 redrawn districts. CA-22, FL-14, FL-22, NC-01, OH-07, OH-09, TX-15, TX-35 have new lines and no official 2024 House result recast onto them, so the model returns no margin. To give a view anyway, we lean a race only when two independent outside signals agree: the latest public poll and the Crystal Ball rating. That gives CA-22, NC-01, TX-15 to Democrats. The other 5 stay coin flips. These leans are analyst judgment, Low confidence by definition, and are kept out of the model totals.

Senate targets

Race	2024 baseline	Candidate	Model margin	Call	Latest poll (comparison)	Ratings	Watch
MN Peggy Flanagan (D) vs Michele Tafoya (R)	D+4.2 2024 president	+0.0 open seat	D+14.2 alt D+15.5	D open Moderate	D+14.0 New York Times / Siena 2026-10-19–2026-10-26	Cook Lean D CB Lean D	—
NH Chris Pappas (D) vs John E. Sununu (R)	D+2.8 2024 president	+0.0 open seat	D+12.8 alt D+14.1	D open Moderate	D+7.0 UMass Lowell / YouGov 2026-10-02–2026-10-19	Cook Lean D CB Likely D	—
AK Mary Peltola (D) vs Dan Sullivan (R, inc.)	R+13.1 2024 president	+15.1 inherited estimate	D+12.0 alt D+13.3	D flip Low	D+2.0 Data for Progress 2026-10-09–2026-10-17	Cook Toss Up CB Lean D	candidate term decides
OH-special Sherrod Brown (D) vs Jon Husted (R, inc.)	R+11.2 2024 president	+10.2 inherited estimate	D+9.0 alt D+10.3	D flip Low	D+3.0 Suffolk released 2026-10-28, field dates not stated	Cook Toss Up CB Toss Up	candidate term decides
MI Abdul El-Sayed (D) vs Mike Rogers (R)	R+1.4 2024 president	+0.0 open seat	D+8.6 alt D+9.9	D open Moderate	D+7.0 CNN released 2026-10-29, field dates not stated	Cook Toss Up CB Lean D	—
ME Troy Jackson (D) vs Susan Collins (R, inc.)	D+6.9 2024 president	-17.6 inherited estimate	R+0.7 alt D+0.6	R hold Low	D+4.0 CNN released 2026-10-29, field dates not stated	Cook Toss Up CB Toss Up	poll disagreescandidate term decidesinside toss zone
IA Josh Turek (D) vs Ashley Hinson (R)	R+13.2 2024 president	+0.0 open seat	R+3.2 alt R+1.9	R open Low	R+2.0 co/efficient released 2026-10-29, field dates not stated	Cook Toss Up CB Toss Up	—
TX James Talarico (D) vs Ken Paxton (R)	R+13.7 2024 president	+0.0 open seat	R+3.7 alt R+2.4	R open Low	R+1.0 Quantus released 2026-10-30, field dates not stated	Cook Toss Up CB Toss Up	—
KS Adam Hamilton (D) vs Roger Marshall (R, inc.)	R+16.1 2024 president	+0.0 no measurement	R+6.1 alt R+4.8	R hold Low	D+10.0 Washington Post released 2026-10-30, field dates not stated	Cook Toss Up CB Toss Up	poll disagrees
SC Annie Andrews (D) vs Darline Graham (R, inc.)	R+17.9 2024 president	+0.0 no measurement	R+7.9 alt R+6.6	R hold Moderate	Even New York Times / Siena 2026-10-19–2026-10-26	Cook Lean R CB Lean R	—

Senate baselines are official FEC 2024 presidential margins. The Alaska (+15.1), Maine (−17.6) and Ohio (+10.2) candidate terms are inherited estimates from earlier races, not new 2026 measurements. Paxton's scandals receive no numerical penalty because no electoral effect has been measured.

How big does the wave need to be?

Each race changes party at the national environment where its 2024 baseline plus candidate term reaches zero. Races near the chosen D+10.0 line are the ones the environment decides; races far from it are decided by the district.

Only ME changes between the two disclosed settings (D+10.0 chosen, D+11.3 raw). Within three points of the chosen setting: MT-01 (D+7.7), NY-17 (D+9.8), ME (D+10.7), VA-01 (D+12.8).

Democratic Senate seats by national environment. The 50–50 tie holds from D+1.4 to D+10.7: losing a Democratic seat takes a much weaker year, while ME flips with only 0.7 more points.

Democratic House seats (220 rated plus toss-up calls) by environment. Control does not depend on the toss-ups; the size of the majority does.

Risk register

Every called race with a warning sign, closest first. The last column is the model's own arithmetic: how much the national environment or a candidate's personal vote would have to move to change the result.

Race	Our call	Warning signs	What would change the call
NY-17 Conley v Lawler	D flip D+0.2 Low	inside toss zone Poll D+2.0, Emerson / PIX11	Flips to R if the national environment is weaker than D+9.8; or if Lawler's personal vote is worth more than 3.7 points (model uses 3.5)
ME Jackson v Collins	R hold R+0.7 Low	poll disagreescandidate term decidesinside toss zone Poll D+4.0, CNN	Flips to D at a D+10.7 environment or stronger; or if Collins's personal vote is worth less than 16.9 points (model uses 17.6)
MT-01 Forstag v Flint	D open D+2.3 Moderate	poll disagreesopen seat: R winner's 2024 vote in baseline Poll R+2.0, RedBox Research / A Fight Worth Having super PAC	Flips to R if the national environment is weaker than D+7.7
VA-01 Taylor v Wittman	R hold R+2.8 Moderate	poll disagrees Poll D+4.0, Washington Post-Schar School	Flips to D at a D+12.8 environment or stronger
MI-10 Hines v Bouchard	D open D+3.9 Moderate	poll disagreesopen seat: R winner's 2024 vote in baseline Poll R+4.8, RedBox Research / A Fight Worth Having super PAC	Flips to R if the national environment is weaker than D+6.1
IA-02 James v Mitchell	R open R+5.6 Low	poll disagreesopen seat: R winner's 2024 vote in baseline Poll D+3.0, Emerson / Poll Vault	Flips to D at a D+15.6 environment or stronger
KS Hamilton v Marshall	R hold R+6.1 Low	poll disagrees Poll D+10.0, Washington Post	Flips to D at a D+16.1 environment or stronger
OH-special Brown v Husted	D flip D+9.0 Low	candidate term decides Poll D+3.0, Suffolk	Flips to R if the national environment is weaker than D+1.0; or if Brown's personal vote is worth less than 1.2 points (model uses 10.2)
AK Peltola v Sullivan	D flip D+12.0 Low	candidate term decides Poll D+2.0, Data for Progress	Flips to R if the national environment is weaker than R+2.0; or if Peltola's personal vote is worth less than 3.1 points (model uses 15.1)
KY-06 Dembo v Alvarado	R open R+16.1 Moderate	poll disagreesopen seat: R winner's 2024 vote in baseline Poll D+6.0, Public Policy Polling for House Majority PAC	Flips to D at a D+26.1 environment or stronger

Part 2 · Digital Atlas context

What the Atlas adds to an election call

Ratings and polls say who is ahead. Digital Atlas says what a district is: the places people use, where they work, how they commute, what housing costs and what public money has actually been delivered. We use that context to explain each seat, test the stories told about it and flag where the arithmetic may be wrong. It never adds a point to a margin.

Moves the margin

2024 resultFEC or official House count

National environmentD+10.0 chosen · D+11.3 raw

Candidate recordmeasured overperformance only

Margin, call and confidence

Explains and stress-tests · adds 0 points

Places10,341,094 in 50 states; 466,754 mapped to exact 2026 lines

Jobs and commutingCensus LODES on 18,520 blocks (AZ-06)

Labor momentum18 FRED state series with receipts

Migration and housingCensus components, ACS rent burden, permits

Promise to deliveryappropriations, contracts, construction

Polls and ratingssponsor, dates and frame disclosed

→

Why it's competitive · what voters face · where the model may be wrong

District fingerprints: exact 2026 lines

Every Atlas place inside each district polygon, grouped into seven categories. Shares are counts divided by all places mapped in the district. The colour scale runs within each column, so a light cell is the highest share of that category among these 18 districts.

District	Schools	Health services	Parks & rec	Manufacturing	Logistics	Religious orgs	Knowledge (IT/cowork)	Places mapped
CA-22	2.80%	0.50%	1.00%	1.25%	1.51%	4.61%	0.30%	15,489
FL-14	1.74%	0.73%	1.31%	0.95%	1.12%	2.60%	1.14%	26,651
FL-22	1.79%	0.66%	1.62%	0.90%	0.95%	2.16%	1.10%	21,730
IA-02	2.75%	1.15%	2.64%	2.17%	1.28%	4.80%	0.40%	25,858
KY-06	2.36%	0.76%	1.24%	1.62%	1.05%	4.65%	0.48%	22,035
MI-04	2.67%	0.90%	1.90%	3.18%	1.30%	4.57%	0.49%	22,547
MI-07	2.47%	0.96%	1.79%	2.06%	0.97%	3.49%	0.58%	21,548
MI-10	1.87%	0.99%	0.84%	4.23%	1.53%	2.05%	0.53%	21,881
MT-01	2.27%	0.64%	2.26%	1.40%	0.88%	2.87%	0.48%	23,511
NC-01	2.27%	0.75%	1.39%	1.29%	0.76%	8.80%	0.19%	25,147
NY-17	2.52%	0.55%	2.25%	1.02%	0.79%	3.52%	0.85%	22,923
OH-07	1.89%	1.01%	1.61%	3.30%	1.79%	2.85%	0.80%	22,572
OH-09	2.27%	0.90%	1.82%	2.51%	1.51%	4.67%	0.42%	22,849
PA-08	2.03%	0.69%	1.77%	1.57%	1.42%	3.43%	0.35%	23,702
TX-15	3.48%	0.71%	1.27%	0.97%	1.61%	4.39%	0.19%	17,750
TX-35	2.19%	0.78%	1.88%	1.50%	1.42%	3.75%	0.26%	19,055
VA-01	1.88%	0.77%	1.09%	0.74%	0.89%	4.10%	0.71%	22,538
AZ-06	2.19%	1.44%	1.72%	0.82%	0.37%	3.01%	0.50%	20,269

Standouts

Category	Highest	Share	Places	vs median
Schools	TX-15	3.48%	617	1.5×
Health services	AZ-06	1.44%	292	1.9×
Parks & rec	IA-02	2.64%	683	1.6×
Manufacturing	MI-10	4.23%	925	2.9×
Logistics	OH-07	1.79%	405	1.5×
Religious orgs	NC-01	8.80%	2,213	2.4×
Knowledge (IT/cowork)	FL-14	1.14%	304	2.3×

How to read it

MI-10 has the highest manufacturing share of the set (4.23%, 2.9× the median), and its race coverage names tariffs/manufacturing as a local issue. NC-01 has the highest religious-organization share of the set (8.80%, 2.4× the median). AZ-06 has the highest health-services share of the set (1.44%, 1.9× the median); block-level jobs data agree, with health care & social assistance its largest employer sector (18.6% of primary jobs). These are counts of places, not employment, spending or votes. They tell a campaign or an analyst which local systems matter in a seat; the model gives them no weight.

Exact point-in-polygon overlay on TIGER 2026 CD120. 466,754 places across 21 districts (17 toss-ups, AZ-06 and three peers).

What voters in these races are talking about

Cost of living & housing24 of 27

Health care & Social Security21 of 27

Energy, utilities & data centers15 of 27

Tariffs, trade & manufacturing11 of 27

Immigration & border7 of 27

Land, water & disaster recovery7 of 27

Ethics & governance5 of 27

War & foreign policy4 of 27

Themes grouped by keyword from the local issues in each race's evidence note; a race counts once per theme.

The sleeper issue: data centers and power bills

Data centers appear as a named local issue in 11 of 27 races: FL-22, KY-06, MI-04, MI-07, MT-01, OH-09, PA-08, VA-01, OH-special, TX, SC. Costs and health care are near-universal; data-center siting, utility rates and grid strain are the most common place-specific issue. This is the clearest next Atlas layer to build: data-center sites, interconnection queues and utility-rate changes by district.

State place economies

Across 50 states, the density of some place types tracks the 2024 presidential margin closely. These are ecological associations between places and states, not between people and votes, and none is a model input.

Pearson correlation between per-capita place density and the approximate 2024 Trump margin, 50 states. Bars marked null (|r| under 0.1) show no linear association: schools (+0.03) and manufacturing (+0.06). The near-zero manufacturing result is evidence against a simple “factory towns vote one way” story.

Pooled place density, blue states = 100

Category	Blue	Red	2024 battlegrounds
Hunting, fishing & firearms	100	157	166
Gas stations	100	154	138
Religious organizations	100	148	136
Manufacturing & metal	100	102	119
Government & emergency	100	125	115
Hospitals & urgent care	100	137	114
Ag & farm supply	100	117	110
Logistics & freight	100	105	98
Schools (K-12 + univ + pre-K)	100	90	88
Nonprofits & social services	100	81	86
Parks, trails & rec	100	75	80
Knowledge economy (IT + coworking)	100	69	73
Transit (bus + rail)	100	53	62

Reading the pools

Counts and populations are summed within each group before dividing, so large states are not averaged with small ones. The 2024 presidential battlegrounds (AZ, GA, MI, NV, NC, PA, WI) carry more manufacturing (119) and hunting and firearms places (166) than either blue or red states: a mixed structure, not a partisan one.

The 50-state aggregate holds 10,341,094 places, 17,296 fewer than the national file (10,358,390). The gap is unresolved and disclosed. Taxonomy v8 has no oil, gas or mining categories.

Worked example: AZ-06, every layer applied

2022 resultR+1.5

→

2024 resultR+2.5

→

2026 modelD+7.5alt D+8.8 · Moderate

Mendoza (D) vs Ciscomani (R, inc.) · Cook Lean D · District poll D+4.5 (super PAC, ±9.1 on the gap) · same lines as 2024

Places in the district20,269Health & Medical is the largest category (14.4%)

Workplace jobs, 2023236,874+9.6% since 2019 · primary jobs, LODES

Live and work inside46.5%133,432 home-to-work flows within the district

Renters cost-burdened52.6%Pima County; Cochise 41.8% (whole counties, ACS 2024)

Largest job sectors inside the district, 2023

Health care & social assistance18.6%

Retail trade11.6%

Manufacturing11.4%

Accommodation & food10.7%

Admin, support & waste services8.9%

Professional & technical services6.8%

Census LODES 8.4 workplace primary jobs on 18,520 whole blocks assigned to AZ-06. Primary jobs are one per worker, not all jobs; 2023 is the latest year.

What the layers showed

Jobs	Workplace jobs grew 9.6% from 2019 to 2023. There is no jobs collapse to explain a flip.
Migration	Cochise gained 1,385 through domestic moves; Pima lost 350 domestically and gained 3,239 from abroad. No single “newcomers” story.
Housing	52.6% of Pima renters spend 30% or more of income on rent. Permitted units fell 11.3% in Pima in 2025 and rose 5.0% in Cochise.
Costs	Groceries are the hardest cost for 30% of likely voters, health care for 16% (super PAC poll). Trump approval in the district: 41.4%.
Delivery	New Douglas commercial land port: under construction; Buffalo Soldier Trail, phase one: design; Pima JTED campus expansion: appropriated; Tucson TARP PFAS pretreatment: operational regional project. Real deliverables in progress; the model gives the incumbent no automatic credit for them.

Bottom line on AZ-06. A two-time Republican winner with a 2.5-point cushion faces a D+10.0 environment. Atlas evidence rules out the decline and migration narratives and confirms affordability as the live local issue. The direction (D) is more certain than the margin: the one district poll has D+4.5, below the model's D+7.5.

For data scientists

Methods, validation and what to fix next

The model is deliberately simple: uniform national swing on an official 2024 baseline plus a measured candidate term. Simplicity keeps it auditable four days out. This tab sets out the arithmetic, how it has scored, and the biases a reviewer should weigh.

margin_D = baseline₂₀₂₄ + environment + candidate_D

House baseline: the district's official 2024 House margin, only where the 2026 lines match. Senate baseline: the official FEC 2024 presidential margin in the state. Environment: D+10.0, a judgment parameter informed by the raw generic-ballot shift 8.7 − (−2.6) = 11.3, which is shown as the alternative. Example: Maine = D+6.9 + 10.0 − 17.6 = R+0.7.

Parameters

Environment	D+10.0 chosen; D+11.3 raw alternative
Toss zone	\|margin\| under 2 forces Low
Candidate term	Overperformance against own party's baseline; weights 0.65/0.35 for two races; one race gets a 50% haircut; open seats get 0
Scandals	No points without a measured electoral effect
Turnout	+1.5 D heuristic documented, never applied
Atlas, polls, ratings	0 points; context and comparison only

Confidence rules

High	High-quality baseline, \|margin\| ≥ 4, a reliable dated market agreeing, no open-seat or new-lines flag. No race qualifies: no market prices were captured.
Moderate	No Low trigger. 11 calls.
Low	Inside the toss zone, a low-quality baseline, a dated rating that disagrees, or a map-expansion state (OH, TX, IA, AK, KS) without market agreement. 8 calls.

How each margin is built

2024 baselineEnvironment (+10.0)Candidate termFinal margin and call

The candidate term changes the winner in 3 races: AK, ME, OH-special. Every other call is set by baseline plus environment.

Backtest: 25 Senate races, 2018 and 2022

20182022

Right direction17/2568% of races

Outside the toss zone15/2268% correct

Median miss5.7 ptsabsolute margin error

The backtest scores the core alone, without candidate terms, using each cycle's own presidential baseline and generic-ballot shift. Misses: 2018 FL, MT, WV, TX; 2022 PA, GA, NV, AZ. All four 2022 misses erred the same way: actual results were more Democratic than the model. That is correlated error, not random noise; the backtest alone cannot say why.

It is Senate-only and in-sample. It says nothing direct about House accuracy, and it is not a calibrated probability. 10 of our 19 current calls sit within one median miss of zero.

Model against the polls

Polls are never inputs, but the comparison is a useful diagnostic. Across the 9 House races with both, the latest poll is on average 3.1 points more Democratic than the model (median +1.8) and points the other way in 5 of 9. In the Senate the average gap is +0.9 (median +0.5), with 2 directional disagreements (KS, ME) and 1 tie.

Race	Model	Poll	Poll − model	Direction	Sponsor
KY-06	R+16.1	D+6.0	+22.1	disagrees	Public Policy Polling for House Majority PAC super PAC
KS	R+6.1	D+10.0	+16.1	disagrees	Washington Post
AK	D+12.0	D+2.0	-10.0	agrees	Data for Progress
MI-10	D+3.9	R+4.8	-8.7	disagrees	RedBox Research / A Fight Worth Having super PAC
IA-02	R+5.6	D+3.0	+8.6	disagrees	Emerson / Poll Vault
SC	R+7.9	Even	+7.9	tie	New York Times / Siena
VA-01	R+2.8	D+4.0	+6.8	disagrees	Washington Post-Schar School
OH-special	D+9.0	D+3.0	-6.0	agrees	Suffolk
NH	D+12.8	D+7.0	-5.8	agrees	UMass Lowell / YouGov
ME	R+0.7	D+4.0	+4.7	disagrees	CNN
MT-01	D+2.3	R+2.0	-4.3	disagrees	RedBox Research / A Fight Worth Having super PAC
PA-08	D+8.4	D+5.0	-3.4	agrees	New York Times / Philadelphia Inquirer / Siena
MI-07	D+5.0	D+8.0	+3.0	agrees	Emerson / Poll Vault
TX	R+3.7	R+1.0	+2.7	agrees	Quantus
NY-17	D+0.2	D+2.0	+1.8	agrees	Emerson / PIX11
MI-04	R+4.8	R+3.2	+1.6	agrees	RedBox Research / A Fight Worth Having super PAC
MI	D+8.6	D+7.0	-1.6	agrees	CNN
IA	R+3.2	R+2.0	+1.2	agrees	co/efficient
MN	D+14.2	D+14.0	-0.2	agrees	New York Times / Siena

Single-district polls carry wide error, and 9 of the 27 latest polls are sponsored by super PACs. 6 of the 10 Senate polls lack published field dates.

Reviewer's bias register

Issue	Evidence	Likely direction	Fix
One shared environment	Every race moves with the same +10.0. A miss in the national number moves several seats together.	Seat totals vary more than independent race errors imply	Simulate with a correlated environment draw calibrated on past cycles
Open-seat baselines	IA-02, KY-06, MI-10, MT-01 use 2024 House margins won by Republicans who are not on the 2026 ballot. The method's open-seat rule (revert to the presidential baseline) cannot run until district presidential results are joined.	Toward R in those seats	Join 2024 presidential results by CD120 district
Inherited candidate terms	AK, ME, OH-special are decided by terms measured in earlier races. Brown's measured edge fell from 14.9 to 7.6 between his last two races (methodology §7).	Either way; brands tend to decay	Re-measure on 2024; publish the no-candidate variant beside the call
Uniform swing	The same swing is applied in an R+26 seat and an R+2 seat.	Over-swings polarised seats, under-swings elastic ones	District elasticity from 2018 and 2022
Senate-only error band	The 5.7-point median miss comes from 25 Senate races.	Unknown for House	House backtest on 2018 and 2022 lines
No market data	All 27 market prices are null.	Confidence capped at Moderate	Capture dated Kalshi and Polymarket snapshots
Redrawn districts	8 of 17 House toss-ups have no comparable baseline.	No call either way; they stay unresolved	Recast 2024 precinct House votes onto CD120

Next steps, by value

Join district presidential results to CD120. Fixes the open-seat baseline, adds a second baseline for every House race and tests the House calls against the methodology's own rule.
Recast the eight redrawn districts. Turns 8 unresolved toss-ups into model calls.
Correlated simulation. Draw the environment once per run, add race noise from the backtest, and report seat distributions instead of single totals.
House backtest. Replace the Senate-derived error band with a House one.
Dated market snapshots. Makes High confidence reachable and gives an independent check.
Atlas energy layer. Data-center sites, grid interconnection and utility rates by district, since this issue appears in a third of the races.

Provenance

Three kinds of number appear in this report and are never mixed: computed (from raw data: places, LODES, FRED, FEC), sourced (dated public facts: ratings, polls, project stages) and model-estimated (the environment, candidate terms and margins).

Generated by build_executive_report.py from race_results.json (schema digital-atlas.midterm-race-results.v1, generated by build_race_results.py using race_model.compute_result). The race results carry SHA-256 receipts for 22 source artifacts:

Source artifact hashes

`election-2026/race_registry.json`	`cc37bd46d05cdc27…`
`election-2026/structural_baselines.json`	`db29f02b585c03f0…`
`election-2026/state_category_distributions.jsonl`	`9c86858713f54ad7…`
`final-assembly/reports/election-context-20261028/election_baseline.json`	`98f73fea4bd41b24…`
`election-2026/sources/labor/ak_ur.csv`	`8c6acbc00fb9bc51…`
`election-2026/sources/labor/ca_ur.csv`	`39c6ef3c85f3c194…`
`election-2026/sources/labor/fl_ur.csv`	`62ac475c336caf6d…`
`election-2026/sources/labor/ia_ur.csv`	`7f65977a8aa92020…`
`election-2026/sources/labor/ks_ur.csv`	`511339d7a90ca5e1…`
`election-2026/sources/labor/ky_ur.csv`	`2f2e0b8a2e598ccf…`
`election-2026/sources/labor/me_ur.csv`	`9850bbf0e42dc596…`
`election-2026/sources/labor/mi_ur.csv`	`83f180f677b5fc44…`
`election-2026/sources/labor/mn_ur.csv`	`f55a7b27ada19b88…`
`election-2026/sources/labor/mt_ur.csv`	`3303675692e12674…`
`election-2026/sources/labor/nc_ur.csv`	`ed3cf860dbc855cd…`
`election-2026/sources/labor/nh_ur.csv`	`bdc00cb3665b1efc…`
`election-2026/sources/labor/ny_ur.csv`	`3bad7790935b9f89…`
`election-2026/sources/labor/oh_ur.csv`	`1ad7be27ecc7da9d…`
`election-2026/sources/labor/pa_ur.csv`	`fb2593dc2ff203de…`
`election-2026/sources/labor/sc_ur.csv`	`7a3ee8dc0cf8258b…`
`election-2026/sources/labor/tx_ur.csv`	`f04d70c597c1fdbd…`
`election-2026/sources/labor/va_ur.csv`	`b6b9935726236e6f…`

Built from the existing reports: Part 1 (battlefield), Part 2 (place economies), Part 3 (Congress), the AZ-06 case study and the methodology. No new research, parameters or model runs were added; the external-lean layer and the sensitivity charts are derived arithmetic.

Disclaimer

Case studies are simple analyses that show what Digital Atlas data can do. Read them on these terms:

Information only. Nothing in a case study is political, investment, financial, legal, betting or other professional advice, and nothing in it recommends taking or avoiding any action.
Not for decisions. Do not use a case study to make voting, campaign, investment, trading, betting, business or any other decisions, and do not rely on it in place of your own research and professional advice.
Not a prediction or guarantee. Calls, scores and estimates come from deliberately simple models with stated assumptions. They are not probabilities or forecasts of any outcome, and actual results may differ materially.
No endorsement. Propheus Pte. Ltd. does not support or oppose any candidate, party, campaign, company or product, and is not affiliated with any of them.
Third-party data. Polls, ratings and other public figures belong to their publishers and are reported as published, with their dates. We do not verify them and are not responsible for their accuracy.
Fixed at a date. Each case study reflects data available on the date it states. It will not be updated and may contain errors or omissions.
No responsibility or liability. Case studies are provided "as is", without warranties of any kind. To the fullest extent permitted by law, Propheus Pte. Ltd. accepts no responsibility or liability for any loss or damage arising from their use or from reliance on them.

Use of this website, including case studies, is subject to our Terms of Service. Read the Terms of Service.