# Polling and financial refresh — October 19, 2026

Both forecasts now use the newly retrieved NYT files and an October 19 polling cutoff. Raw editions, revisions, previous forecasts and per-step logs are preserved. Forecast settings and feature-admission rules remain unchanged; moving the forecast date also reruns historical comparisons at the matching horizon.

NYT source changes: Senate **35 new candidate rows / 9 new poll groups**, with 7 new admitted questions and 2 explicitly excluded questions. House **19 new rows / 4 new poll groups**, plus 2 revised rows. A source poll group can contain multiple questions or candidates; source additions are not all automatically eligible polls. Date, nominee, hypothetical-matchup and clustering rules still apply.

## Forecast movement

- House expected Democratic seats: **231.70 → 231.40**. Democratic majority chance: **85.2% → 85.0%**.
- Senate expected Democratic-caucus seats: **50.01 → 49.92**. Chance of 51 or more: **42.7% → 41.2%**.
- Current generic ballot: **D+8.14** points.

## Financial progress

Matched cutoff-valid Q3 reports increased from **60 to 136 mapped candidate-in-election instances**. Comparable quarter features cover **60 historical competitive races and all 35 current races / 71 candidates**. Historical 2020/2024 report gaps are filled. One Gary Peters financial identity mismatch was corrected. Current financial features retain their October 18 disclosure cutoff.

The refreshed standalone tests and the new 192-combination financial/fundamentals study do not justify an added financial weight. The combined study uses a frozen earlier polling context; its best retrospective margin MAE is 4.100 points, versus 4.088 for that baseline. An earlier-only variable selection policy also worsens error. The earlier combination pages retain their frozen earlier cohorts. See the separate [financial update](../financial_refresh_2026-10-19/RESULTS.md).

## Sources and downloads

- [NYT source checks and immutable editions](polling_source_checks.json)
- [New Senate questions: admission/exclusion audit](new_senate_question_review.csv)
- [Every race's forecast movement](forecast_changes.csv)
- [Independent refresh validation](validation.json)
- [Build steps and logs](rebuild_steps.json)
- [Final completion checks and notebook recovery](completion.json)
- [Browser-review limit](browser_review.json)
- [Current Q3 individual fundraising](../financial_refresh_2026-10-19/current_q3_individual_fundraising.csv)

All numerical checks passed. Live browser inspection was blocked because the app browser prohibits the local-file protocol; no visual or interaction pass is claimed. Static checks confirm that sort headers, coverage filters and interval definitions are retained. New polling and updated recency/horizon drive the forecast changes; this update does not deploy the experimental interval or financial specifications from the prior review.