# Financial report update — October 19, 2026

Recovered cutoff-valid Q3 reports for all **136 mapped candidate-in-election instances**, up from **60**: **76 additional matched records**. This closes the mapped 2020 and 2024 Q3-report gaps. It is not coverage of every Senate candidate nationwide.

The new retrieval also exposed a supplied Gary Peters candidate-ID mismatch. His financial join is corrected to S4MI00355 / Peters for Michigan C00437889; original election source records remain preserved. Candidate names, state, office, party affiliation, committee and reporting dates were checked, including Minnesota DFL and Al Gross's independent affiliation.

Comparable full-quarter race pairs: **2016: 8; 2018: 11; 2020: 16; 2022: 13; 2024: 12** — 60 historical competitive races. Current quarter coverage is **35 races and 71 candidates**, including all three modeled Montana candidates. Cumulative-period historical coverage is 7/10/16/13/12, and current coverage is 35 races.

The financial features retain the **October 18 cutoff**; these source downloads were checked October 19. Only reports/amendments received by each historical forecast cutoff enter the tests. Full-quarter amounts use reported July–September activity or reconciled September-minus-June totals. Missing values are never treated as zero.

## Initial refreshed tests

All three standalone individual-dollar definitions worsen average full-quarter margin error over 2018/2020/2022/2024. The polling baseline averages **3.95 points**; individual-dollar share gives **4.28**, the dollar gap **4.31**, and log ratio **4.28**. Dollar amount helps in 2024 but does not hold up across elections. Cumulative definitions also worsen their comparable averages. This does not establish a causal effect of money, nor does it rule out useful combinations with other factors.

No new financial forecast adjustment is enabled. A fresh, separately saved combined test now covers 41 comparable races in 2020/2022/2024, with 2016/2018 used for earlier training. Every one of 192 factor/financial-definition combinations uses the same race cohort. The best retrospective fixed-25% blend scores **4.100 points**, versus **4.088** for the frozen polling baseline; it omits fundraising. Selecting factors, ridge and blend using earlier validation elections scores **4.228 points** and also does not improve the baseline. Only 2024 has enough earlier validation environments to test a learned selection; the earlier two tests fall back to no adjustment. There is no established significant financial benefit.

See the [expanded combined-financial study](../../model/extensions/financial_refresh_2026-10-19/RESULTS.md), including 15 independent arithmetic, chronology and preservation checks. The old combined-fundamentals pages remain frozen; these new results use the larger financial sample and their own saved inputs.

## Downloads and checks

- [All current Q3 individual totals and shares](current_q3_individual_fundraising.csv)
- [Updated standalone comparison summary](comparison_summary.csv)
- [Candidate-identity checks](candidate_identity_checks.csv)
- [Report reconciliation and source filings](report_reconciliation.csv)
- [Independent validation receipt](validation.json)
- [All financial race features](../../model/extensions/senate_candidates/individual_dollar_race_features.csv)

Individual dollars include itemized and unitemized contributions, not outside spending, loans or all receipts. These are dollar amounts/shares, not unique donor counts. In-state giving still needs cutoff-dated transaction reconstruction. All downloaded report files and earlier derived datasets are preserved.