conclusion:

1.

To calculate the ratio we don't need any knowledge on the Model
only about the two policies we use.

2.

The ratio is $(\frac{D_{2}(X)}{D_{1}(X)})F(x)$ is the case of Importance Sampling.

3.

1+2 imply we can use samples from one policy to calculate samples on another policy.

4.

conclusion 3 explain why Q-learning can work.

5.

The Variance must be limited to avoid errors.