How max_seq_len param impacts custom LSTM implementation

I am creating a custom LSTM model for PPO to handle a time sequence of say 10 observations. Doing that as advised in the docs as per below:

from_ = observation_time_sequence_len - 1
self.view_requirements[SampleBatch.OBS].shift =
“-{}:0”.format(from_)
self.view_requirements[SampleBatch.OBS].shift_from = -from_
self.view_requirements[SampleBatch.OBS].shift_to = 0

Doing that I do not have to add a time dimension in the forward method.

How does the RLLIb max_seq_len param impacts this? I do not quite understand how my observation_time_sequence_len could override or at least work in concert with the max_seq_len.

Seeing some different results changing the max_seq_len value.