House: a conditional model
Current-boundary 2024 presidential margins + retained historical House residuals + an assumed 4.5-point Democratic margin improvement. All 435 current-map presidential inputs were freshly checked and are unchanged. The national shift is assumed, not estimated from early ballots. Residuals are capped at ±12 points, generally retained fully on unchanged maps, half in changed-map states and zero for selected departures.
Historical House margins have not been fully reallocated to current boundaries. State-level retention is not a measured district-overlap or candidate effect. Candidate verification covers 175 distinct districts, adding 68 major-party lineups in Florida, North Carolina and New York. Florida’s tracker is an unofficial reference; minor parties and write-ins are not exhaustively checked. California and Texas checks are carried from October 22. Nationwide verification remains incomplete.
The inherited majority threshold is just above +3.85 points. At +4.5, presidential results alone give 217 D / 218 R; removing selected former Democratic incumbents’ residuals in PA-07, PA-08 and CO-08 gives 218 D / 217 R. Seventeen additional Florida/New York departure resets change no allocations across the five tested shifts. Resetting a whole residual also removes opponent effects; these are sensitivity cases, not measured candidate corrections or confidence intervals.
CA-06 is Pan (D) versus Kiley (No Party Preference); central allocation D, with separate caucus treatment required for a Kiley win. CA-40 is R versus R; CA-04/07/11/12/14/29/34/37 are D versus D, so hypothetical D–R margins there are not candidate-margin predictions. Florida incumbent district-number changes illustrate why same-number historical residuals are not personal candidate advantages. The slider is conditional sensitivity, not a new independent forecast.
Senate: judgment, not a probability model
14 D and 21 R allocated wins plus 34 Democratic-caucus and 31 Republican seats not up give 48 D / 52 R. The D total includes two existing independents; Nebraska’s Dan Osborn is not automatically counted as a Democrat. A fresh certification of every Senate nominee remains incomplete.
North Carolina, Georgia, Michigan and New Hampshire remain D leans. Ohio special and Alaska are toss-ups allocated R for arithmetic; Maine is near a toss-up allocated R. Nebraska and Iowa lean R; Texas, Kansas and Florida special favor R. Historical results and candidate effects underpin many allocations where early records cannot identify direction.
Ohio: a roughly R+11.2 presidential baseline, the assumed 4.5-point D shift and 7.6 points of retained Brown overperformance yield D+0.9; six points yield R+0.7 and four yield R+2.7. Full persistence is not established. Maine’s near-toss-up judgment assumes substantial decay in Collins’s approximately 17.7-point 2020 advantage relative to the presidential margin. Alaska depends on Peltola’s crossover and Sullivan’s incumbency.
Ohio and Alaska flipping yields 50–50 with Republican vice-presidential control; adding Maine gives 51 D / 49 R. North Carolina reverting to R alone gives 47 D / 53 R. These are stress scenarios, not calibrated probabilities.
What goes in
The completed October 23 report supplies this snapshot using actual voting records and historical election results. Polls, exit polls, prediction markets, external race ratings, fundraising and pundit consensus are excluded. Party registration and primary history are not candidate votes. No reported 2026 general-election candidate-vote totals are present in these records.
UF’s common file was modified October 23 at 4:32:01 p.m. Pacific. Forty positive rows are dated October 23; Colorado, North Dakota, Pennsylvania, Rhode Island, Washington and West Virginia October 22; Mississippi October 19. Alabama, Missouri, New Hampshire and New York are unavailable, not zero. California’s party coverage expanded to 99.8%; Idaho has 59.0%. Minnesota’s increase includes a week of reporting catch-up.
Wisconsin: UF reports 377,058 (228,162 mail plus 148,896 in person), while another tracker reports 525,954 by adding 148,896 to 377,058 labeled mail. The difference equals the in-person component. The state report was inaccessible, leaving possible overlap unresolved. UF’s value is retained for consistent aggregate accounting; Wisconsin is excluded from turnout-direction inference, including yesterday’s apparent +6% matched pace.
Supplemental universes stay separate. Nevada’s October 23 5 a.m. accepted total is 288,006, versus later UF 317,012. Arizona’s coherent return series is 843,454 with county cutoffs October 16–23; UF uses its earlier 759,605 snapshot. Georgia Votes reports 1,467,780 at 09:24, Civic Forge 1,467,774 dated accepted records, and later UF 1,613,269. Different cutoffs and undated records prevent combining them; pending, rejected and cancelled ballots are excluded.
Georgia’s improved in-person comparison matches 158 counties excluding Gwinnett and ten calendar days relative to Election Day: October 13–22, 2026 against October 18–27, 2022. The 2026 count is +23.86% versus 2022, −37.22% versus 2024 and +85.57% versus 2018. Registration grew 4.98% versus 2022. Hours, sites and voting-method substitution remain uncontrolled; 2018 is descriptive only, with 96 unmapped records excluded. No fully access-matched 2018 or nationwide historical series is claimed.
Georgia’s latest-primary-history cohorts use cutoffs through May 2026 versus May 2022, unlike the prior fixed historical cutoff. Changing primary contests, crossover and cohort membership prevent interpreting their shifts as candidate-vote swings. All imputed Gwinnett comparisons and candidate scenarios are excluded. Maine, Clark-mail and Oregon comparisons are stale; Pennsylvania’s ballot-status definitions differ. Unavailable mode cells are not proof of zero ballots. Requests and issued ballots are not added to votes.
Common data: University of Florida Election Lab. Arizona: Derek Weech, Stealth Analytics. Georgia: Civic Forge Solutions analysis of state voter records, CC BY 4.0. Geography: Natural Earth. House fills show modeled seat splits within state outlines, not district boundaries.
Trail & coast
