The order of tokens is set by the first serial filter encountered as the tokens pass through the pipeline. So if the very first filter is serial, then the token generation order is used. All subsequent serial filters process the tokens exactly in that order; and you should do nothing for that, except for properly classifying filters as serial or parallel.

Is this level of explanation enough, or you would like to see more details?

By the way it is said in the tutorial that

in the current implementation the is_serial parameter is ignored for the first stage, though this may change in later versions".

This is not true anymore; in the latest development updates the first stage will recognize the filter type like any other stage. The documentation will be soon updated.