The Pipeline Browser down the left is the whole mental model. Your file sits at the top; every filter you apply hangs beneath whatever was selected when you applied it. Select a different node and the next filter branches from there instead. Delete a branch and the ones above it are untouched.

The practical version of this: if a dataset is slow to interact with, apply Extract Subset or a Threshold first and build your pipeline against the smaller version. Swap the input back at the end once the pipeline is right. This is the single biggest speed-up available to a new user.

Trace a session, save the script, then run it with pvpython on the machine holding the data. No window, no clicking, and it produces the same images every time. Save the session itself as a .pvsm state file if what you want back is the interactive setup rather than a batch job.