# September fundraising update — October 21, 2026

The local House and Senate forecasts now use the reviewed Senate total receipts through **September 30** for **24 matched candidate pairs**. Polling remains the **October 19** snapshot. Each simulation keeps the existing polling correction, candidate history, scandal settings and shared national/state/regional errors. The fundraising formula and its three-point cap are unchanged. Georgia runoff frequency and the resulting extra runoff uncertainty are recomputed with the updated race estimate.

| State | Democrat | Previous margin | Updated margin | Previous win chance | Updated win chance |
|---|---|---|---|---|---|
| AK | Mary Peltola | R+1.60 | R+1.56 | 38.77% | 39.02% |
| IA | Josh Turek | R+3.02 | R+2.64 | 27.67% | 30.25% |
| FL | Angie Nixon | R+8.25 | R+7.03 | 6.09% | 9.03% |
| SC | Annie Andrews | R+4.90 | R+3.18 | 19.13% | 28.32% |
| KS | Adam Hamilton | R+2.08 | R+1.58 | 34.70% | 38.25% |

The shared-error projection gives Democrats a **53.2%** chance of at least 51 Senate seats, up from **51.7%**. Newly elected independents are counted separately. House race forecasts are identical to the prior run; the updated Senate outcomes change the joint chamber probabilities.

## What passed

- All 12 rebuild and model-validation steps passed. Four cases each contain 100,000 joint draws.
- All 31 directly compared Senate results match the independent September sensitivity calculation; the other four unpolled fallbacks are unchanged. All 435 House rows match the prior forecast exactly in all four cases.
- Historical calibration data, polling inputs, candidate-history policy, scandal policy and error settings retain their earlier checksums.
- The four public FEC summaries for Peltola, Sullivan, Turek and Hinson pass 20 accounting equations to the cent. Their receipts, itemized contributions, cash and transfers match all 16 corresponding stored amounts.
- Sullivan's six-year election-cycle receipts bridge exactly to the two-year input after subtracting the prior-cycle balance. Election-cycle totals and two-year totals are kept distinct.
- All 272 Senate poll clusters match the original NYT question records and have archive-created dates no later than the snapshot date. These timestamps screen for later additions; they do not independently prove first public release.

The historical September comparison covers 142 paired races across six evaluation cycles. Mean absolute margin error moves from 4.891 to 4.857 points. That small improvement is statistically uncertain (cycle sign-flip p=0.406); competitive-race error is slightly worse. The financial adjustment remains provisional.

## Filing reconciliation still open

The official FEC linkage file maps the 48 candidates to 61 authorized committees. The query lists **333** reports; **100** were obtained before the FEC returned a retry window, plus two previously cached Q3 summaries. All **102** retrieved report cash equations balance. The remaining **231 query reports** and their amended-filing chains have not been fully checked. Missing reports are not assigned zero receipts, and the authoritative candidate totals remain in use.

[Candidate and committee coverage](filing_coverage.csv) · [Partial filing checks](partial_filing_reconciliation.json) · [Four public summary checks](public_summary_reconciliation.json) · [Sullivan period bridge](../review_2026-10-21/finance/period_bridge.json)

## Dependency and file cleanup

Removed 17 inactive project-only browser test profiles, freeing **65.6 MB**. Source downloads, screenshots, model inputs and prior audit records were preserved. Duplicate source/pipeline/archive copies were retained where referenced. Active validators now check the selected source period and pinned inputs instead of assuming June forever. Preview links now open the current forecasts, and the downloads point to the active September data.

[Cleanup record](dependency_cleanup.json) · [Duplicate dispositions](duplicate_dispositions_after_cleanup.json) · [Polling source check](source_availability.json)

## Files and checks

[Senate forecast](../../model/Senate%20Model.html) · [House forecast](../../model/House%20Model.html) · [Financial report](../../model/extensions/provisional_finance/Total%20Receipts.html) · [Research overview](../../model/extensions/Research%20Overview.md)

[Independent numerical checks](numerical_validation.json) · [Rebuild steps](steps.json) · [Active matched totals](../../prepared/fundraising/2026-09-30/senate_pairs.json) · [Fixed financial fits](../../prepared/fundraising/2026-09-30/senate_fixed_fits.json)

## Final page checks

Both local forecasts pass 337 interaction checks. The local website presentation passes 141 runtime and saved-data checks. Source and download links resolve, including the September receipt and matched candidate totals. These are script-based checks; a fresh visual browser review was not performed.

[Page interaction checks](validation.json) · [Website presentation checks](presentation_validation.json) · [Source-link checks](source_link_validation.json)

Publication to the existing website was explicitly approved on October 21 after the local review. The earlier local verification receipts retain the publication status at the time they were written. Filing reconciliation remains partial as described above.