I think this system looks less naturalistic. First, is an ACC/ERG case attested? It does look strange to my eyes. If the case marker has different allomorphs based on animateness why not call it two different cases. They serve to different functions after all.
It does make sense though that the case marking of P and A interacts. Martin Haspelmath calls this a scenario in his more recent work.
.
I don't think that marking an A that is animate with the ergative makes a lot of sense. Usually A is marked with an ergative case if it is inanimate. This is because animate nouns are 'expected' to occur as an A and imanimate nouns as a P.
Also, you have included two contradictory statements about the case marking of P. Is it generally dependebt on the case of the A? Or is it also dependent on the animateness of P?
Trying to reformulate what you said without the stuff that I did not find naturalistic.
S is marked with a nominative/absolutive (n/a) case in all contexts.
A is in the ergative case if
a) the verb is in any perfective construction and
b) the A is inanim,
elsewhere A is in the n/a case.
P is in the n/a case if
a) the verb is in any perfective construction and
b) A is inam
c) P is inam,
elsewhere P is marked with the accusative case.
This would mean you have nom-acc alignment in non-perfective constructions. In the perfective aspect scenario A=inam and P=inam you would have erg-abs alignment. In the perfective scenario A=inam, P=anim you would get a tripartite alignment. This looks plausible to me. Also, it's nice to see a fresh take on split-alignment languages