Meaning of bootstrapping residuals

What does it physically mean when I bootstrap residuals in a regression problem. Like when I bootstrap residuals and create synthetic data does it mean I am hypothetically doing some experiments(virtual experiments which are represented by synthetic data). Or does it mean I am just randomly resampling from residuals. What is the physical insight into doing bootstrapping.

One the other note, what would it mean when I bootstrap pairs.


Less is more. Stay pure. Stay poor.
My meagre interpretation is that bootstrapping can be conducted for many reasons and in particular can provide more robust measures or intervals. So yes, resampling over and over again to provide varying samples that may be more reflective of the originally sampled population. So, if your 20% sample was not reflective of the total population the random resampling based measures derived may better reflect the real population or provide more cautious estimates.