Until dark matter can be directly, or indirectly but consistently detected (e.g. we can take a bunch of dark matter and move it around, if it doesn't move it is a property of that particular region of space, not something contained in it), dark matter stays as an abstraction that helps our formulas to explain, pardon, model gravitational interactions.
That is, now you can either consider it an as yet undetected physical object, or the rationalization of an error, as you prefer, and orient your own research accordingly.
I point this out because "our models do not match our observations" can be either resolved by "therefore there is something more to be modelled" which in this case implies the dark matter hypothesis, or "therefore our models are wrong/too general/too limited", which would be strange but not impossible. Even after modeling every single past present and future aspect of reality, you cannot claim you completely know it, since you are speaking from the inside of it. It would be like testing every I/O combination in a unit test and then proclaiming you have achieved 100% code coverage. Those are two different things on two different levels.
Many scientists know this, not all of them unfortunately.