Written on 22 October from the forecast and the coverage as they stood at the end of 16 October.

Money dominated the day's coverage. NBC News reported that Democratic Senate candidates were outraising their Republican opponents across the map. ABC and NBC both covered James Talarico's record quarter in Texas. Axios reported that a senior Trump campaign operative had been brought in to help Senator Roger Marshall in Kansas.

The forecast moved the other way. The chance of Democrats winning the Senate fell from 58% to 56% over the day, and the expected Democratic seat count slipped from 51.1 to 50.9.

Why fundraising is a small input

Fundraising does enter this model. It is one of the "fundamentals", the non-polling information that estimates a race before anyone surveys it, along with the state's partisan lean, incumbency and the national environment. Money raised from a candidate's own state counts five times as much as money from outside it, because local donors are a better signal of local support.

But by mid-October the fundamentals are mostly a backstop. In a race polled as heavily as Texas, which had 26 polls in its average that day, the polls carried 71% of the weight. Talarico's haul is real, and it probably reflects the enthusiasm the polls are also measuring. It cannot add much that the polls have not already captured.

Kansas, where history still has a vote

The day's largest move among competitive Senate races was in Kansas. The margin went from about 1.7 points to 3.6 points in Marshall's favour, and the chance of a change in party fell from 39% to 28%.

The reason is how little polling Kansas had. With four polls, the polls carried just under half the weight in the forecast. Their average had the race almost exactly level, about half a point apart. The fundamentals, built mostly from how Kansas has voted in recent elections, put Marshall ahead by almost nine. The forecast sits between the two.

That is deliberate. Four polls can be four views of the same mistake, and states with a long record of voting one way have a habit of drifting back to it late. If more polls show a level race, the polls will take over. Until then a forecast should not trust four surveys over the state's voting record.

Elsewhere

Ohio moved slightly towards Sherrod Brown, to a margin of about 3 points, and Maine slightly towards Susan Collins, though Troy Jackson still led there by about 4 points. Texas edged towards Ken Paxton, to a margin of under a point, which in a race polled this often is well inside the noise.

The House moved a little the other way, from 90% to 91% for a Democratic majority. On a day when one chamber became slightly more likely to flip and the other slightly less, the clearest lesson was how much each race still depends on how often it has been polled.