Show the evidence.
Respect the uncertainty.

Evidence cutoff: November 3, 2026, 8:00 a.m. EST. Numerical forecast benchmark: November 2 historical entries. Nominee snapshot: October 8. The research covers 435 voting House districts, 35 Senate elections, and 65 Senate holdovers in control arithmetic.

The source says historical series were retrieved later. Publisher revisions cannot be ruled out. No election returns or November 3 forecast entries were used as cutoff inputs in the source research.

The national probabilities and projected margins are attributed to Decision Desk HQ (DDHQ) in the source report. They are not a newly trained Election Outcomes Lab or Bellelytics national model. The earlier Pennsylvania exploratory analysis is separate.

External source pages may now show newer information. Links are a source register, not proof that the displayed snapshot has been independently reproduced here.

The workbook uses these report conventions: Toss-up when the leading group is below 60%; Lean from 60% to below 75%; Likely from 75% to below 95%; Strong at 95% or above.

These are not Cook Political Report categories, guarantees, or measures of predictive accuracy. Cook ratings cited by the report are a separate qualitative cross-check.

The October 8 roster retains every nominee entry in the source workbook, including minor-party candidates and additional same-party nominees. Later withdrawals were not exhaustively audited.

Provider probabilities apply to selected modeled party groups. An unpriced Democratic or minor-party contender is not assigned a real-world zero chance. New independent wins are not assumed to join the Democratic caucus. Ranked-choice and same-party races retain their source caveats.

Published chamber-control probabilities are preserved, not recalculated by multiplying race odds. Expected seats are sums of marginal probabilities and may differ from separately published projected-seat figures.

Margins retain their source percentage-point meaning. No confidence intervals are invented. The upset screen uses a favorite at 65% to 80% and a modeled runner-up at 20% or above; it does not claim an independent forecasting edge.

The build checks coverage, unique race identifiers, source dates, nominee counts, and reconciliation of modeled probabilities. This verifies the site's handling of the supplied data, not the accuracy of the underlying forecasts, poll claims, nominee lists, or historical calibration.

Provisional prototype weights in the source are not treated as a fitted nationwide model. No new economic, demographic, or digital-sentiment effect has been calculated for this website.

The workbook describes the compilation as private research and states that it does not license commercial redistribution of provider databases. The public launch therefore excludes the complete race-level forecast dataset and original source files.

The public site contains selected attributed findings and the nominee directory. Full files are supplied separately for local internal review. Any broader publication should be checked against the applicable provider permissions. Account-specific prior-session links are omitted from this public source register.