and isn't synonymous with 'confidence.' The 95% confidence level means that it would be rare for the true number to fall outside the number +/- the MOE. The smaller the sample size, the larger the MOE and vice versa. Although larger sample size doesn't always mean more accurate until the pollster uses some means to obtain a representative and random sample of the whole population. Easier said then done.

Using an example -- the last 2016 Quinnipiac FL poll.

Clinton 46%
Trump 46%
MOE 3.3%

Actual results -
Clinton 47.8%
Trump 49.0%

The actual results were within the 3.3% MOE. So, the statistical confidence level was accurate.

and yet they started projecting Biden the winner even before he entered the race. But boy did they keep people distracted for those 2 years.

What is the purpose of this information? Does anyone believe that Biden would let either Warren or Bernie into his cabinet?

I've willingly partaken of was for the Minnesota Orchestra, years ago. I suggested that they play a symphony by Franz Schmidt for the upcoming season, and lo and behold, they included his 1st Symphony, so we had to go.

I don't need no stinking polls.

What is polling, but the commodification and exploitation of the public's opinion for profit, not a public good. At least imho.

Considering the sobering results of the Martin Gilens and Benjamin I. Page study, "Testing Theories of American Politics:Elites, Interest Groups, and Average Citizens", we see that when public opinion comes to influencing actual public policy of our government...

"Multivariate analysis indicates that economic elites and organized groups representing business interests have substantial independent impacts on U.S. government policy, while average citizens and mass-based interest groups have little or no independent influence. The results provide substantial support for theories of Economic-Elite Domination and for theories of Biased Pluralism, but not for theories of Majoritarian Electoral Democracy or Majoritarian Pluralism."

snip

When a majority of citizens disagrees with economic elites or with organized interests, they generally lose. Moreover,because of the strong status quo bias built into the U.S. political system, even when fairly large majorities of Americans favor policy change, they generally do not get it.

A perfect example issue is MedicareForAll.
According to Newsweek.com: (4-24-20)

A newly released poll shows that 69 percent of registered voters support Medicare for All, a plan which would create a national health insurance plan available for all Americans.

The poll also showed 46 percent of Republican voters supporting Medicare for All alongside 88 percent of Democrats and 68 percent of Independents.

A clear majority of Americans support the idea of a Medicare For All type system. Neither Trump nor Biden support what clearly a majority of Americans want in their healthcare system.

I'm like, why should they (pollsters) get paid for my opinion that, in the grand scheme of things, really doesn't matter anyway? Why should I give them free, raw materials (my opinion) for their faulty questions (based on linguistic gymnastics), and ultimately worthless products, polls & commentary when public opinion has no influence on public policy?

It generally goes like this...
Me - We live in a capitalist society correct?
Pollster - Well, yes
Me - You get paid to solicit my opinion, right?
Pollster - Well, yes
Me - You want my opinion, pay me. That's how capitalism works, is it not?
Pollster - WTF? (This guy's nuts!) click.

to include the 2016 polls and assume that the 95% confidence reflects the final results.
There were almost 10% of undecided voters in 2016 because they disliked both major party candidates (I include voters who thought they would vote for third-party candidates as undecided). At the last minutes. many of them will vote for one of the two party candidates anyway, usually for the one they know less and therefore hate less.
The state polls are even worse with some 12% undecide.
Therefore, the polls with high percentages of undecided have built-in error as predictors for the future since we're almost guaranteed that the future will change. The confidence level and margin of errors are supposed to apply to the polls at the moment they were taken to reflect the population at the time. If hundreds of polls were taken at the same time, then 95% of them will fall within the margin of error of the average.
If the study removes all polls which have a more reasonable percentage of undecided, like under 5% then I bet it will fare better than 60%. But, fundamentally, they are wrong to assume that there are no major events between the time the polls are taken and the future it's supposed to predict that might change what people think.