Read it again any time, including while you work through the exercises.
Toss a fair coin 100 times. Expecting 50 heads? You might get 47, or 53, or 55. Do it again: yet another result. This phenomenon is called sampling fluctuation: from one sample to the next, the observed frequency moves around, even though the true probability p does not change.
This fluctuation is not a flaw: it is the signature of chance. The good news is that its SIZE is predictable. For a sample of size n, the observed frequency f falls, in about 95% of cases, inside the fluctuation interval: from p minus 1 over square root of n, to p plus 1 over square root of n.
Example: fair coin (p = 0.5), sample of n = 100 tosses. Square root of 100 = 10, so 1 over square root of n = 0.1. Interval: from 0.4 to 0.6. Getting between 40 and 60 heads is unremarkable. Sample size is decisive: with n = 400, the margin drops to 0.05, and the interval tightens: from 0.45 to 0.55. The bigger n is, the less it fluctuates.
This interval gives you a DECISION tool. Someone claims "my coin is fair", you toss it 100 times and get 65 heads, so f = 0.65. But the fluctuation interval for p = 0.5 runs from 0.4 to 0.6, and 0.65 is outside. Such an extreme result is rare if the claim is true: you have a good reason to doubt it, with a risk of being wrong of about 5%.
Remember the key distinction: the observed frequency f is a fluctuating estimate, not the true value of p. A poll showing 52% does not prove that p is exactly 0.52. All of statistical intelligence lives in that nuance, and you are going to learn to handle it.