I deleted my comment earlier, because this really isn’t the important part of it. But I read real-time graphs for a living, and often have to make decisions when the trend isn’t 100% known. I saw no shenanigans in the math. The math isn’t the important part.
Usually on a trend line like this, there’s some sort of weighted moving average being applied. Sometimes it’s a simple “average of the previous 20 points”, sometimes more recent points are weighted more heavily, such as an exponentially-weighted moving average, or EWMA. To my eyes, this trend looks like an EWMA.
I find it easier to see from other angles sometimes:
Two red dots on the upper-left leverage the red line off to the left. Most of that jump is hanging on them.
By contrast, the blue dots don’t jump as far to the left. Biden also had an uptick, but not nearly as far.
There is also a greater proportion of blue dots to the right side of the blue trend than there are red dots on the right side of the red trend.
The math looks clean to me. But the math doesn’t mean much.
There is a lot you can argue about here that’s way more productive than this. For example: polls always depend on getting a representative sample. That’s been hard since smartphones came out; in other words, the problem is over a decade old and there’s no solution in sight. Poll results are skewed heavily towards people they can reach; people who answer calls from people they don’t know, which pretty much excludes Gen Y and younger.
