Ah ofc.. theres always 16 new created samples in case of a 44.1k or 48k signal.
Even mojo creates that samplingrate internally i guess.
Leaves it difficult for me to grasp how past and future samples further than the 1st side ones can make the right 'predicion' as music is so random.. or are you looking to obvious non signal dependant aspects of original samples?
I think the challenge here is that the Nyquist-Shannon sampling theorem is not intuitive. So it's often hard to wrap our minds around it other than to "trust the math".
This is how I try to conceptualize the math. The problem we often have is that the digital samples we have were taken from the original analog waveform.
So if we have two sequential samples and we are trying to reconstruct the analog waveform, intuitively we would say, well, technically there is an infinite number of ways to connect these two dots so anything goes. Moreover, why would the samples occurring 1 second before these two samples or 1 second after these two samples have anything to do with the "correct" analog waveform between these two samples.
But in reality, that's not true because your analog waveform first of all is bandwidth limited to 20Hz to 20kHz. So while technically, there are seemingly still infinite ways to connect the two samples, you're restricted to waveforms that would be part of a 20Hz-20kHz waveforms.
Now let's first imagine we are dealing with a pure tone at a specific frequency. How do we know whether it's a 20Hz tone or a 20kHz tone? Well, the samples before and the samples after would tell us.
This is how I conceptualize why the longer the tap length, the more accurately we can re-construct the original analog waveform. It's still very non-intuitive because you can always increase the tap length so at what point, e.g. 1s, 5s, 10s? does the samples from a long time ago or a long time later no longer influence the analog waveform reconstruction in-between those two specific samples?
But so far, I definitely prefer 1 million taps of WTA over shorter WTA filters over other non-WTA filters.