Polls have a margin of error and election results have generally been within the 80% confidence interval 80% of the time.
It is true that when there are less polls (like in special elections) it’s harder to get an understanding of the state of the race.
I prefer Micromamba since it’s faster at solving environments.