I'm contemplating not using Trafalgar anymore in my model. On the one hand, I don't want to be biased and it's important to include outliers in both directions. On the other hand their polls are just stupid.
On the other hand, their polls were pretty accurate in 2016. Even if they are off by anywhere from 2 to 4 points, they are still better than Quinnipiac polls. Therefore, I recommend you keep them.
I mean, I personally think they get too much credit for being right in 2016. They largely tend to be quite bullish on Rs relative to other polls. In 2016 where Rs widely outperformed polls, especially in the rust belt, they look great, but in 2018, outside FL, they didn't do very well. Since then, the gap between them and other pollsters has become even more extreme. I think I'll keep them, to pull the model a bit more towards Trump and increase uncertainty, as obviously, this cycle has had so many twists and turns