It needs to be falsifiable not necessary to be in a direct or easy way. We still currently have a large portion of the dark sector parameter space unexplored. I would not call that surprisingly small [1].
Also there are many models of dark matter that was excluded by experiments already. I would describe falsifability as a problem fpr MOND models.
[1] https://pdgweb.lbl.gov/2024/reviews/rpp2024-rev-dark-matter....
I'm not sure MOND is harder to falsify than dark matter. There have been some recent observations looking at wide binary stars. Throw as much money into falsifying MOND as we have into trying to find dark matter particles, and we can compare!
And please don't compare that with dark matter because it has very well integrated framework with particle physics. It does not come merely as a way to only explain galaxy rotation curves while having problems with any larger structures.
Making dark matter fit observation involves halos and cusps and getting the distribution just right so that the Tully Fisher relation (which is about the luminous matter) still holds.
As a layman, I find MOND involves considerably less fitting!
Tully-Fisher - dark matter fails to explain the Tully-Fisher relation naturally, and requires a lot of tweaking and feedback effects (arbitrary parameters).
Rotation curves - recent empirical observations shows galactic rotation curves remain flat far beyond what dark matter can explain [1]
Falsified - we've been trying to find dark matter for a long time and failed. The window of possible candidates for dark matter is now surprisingly small.
I am not claiming that dark matter is wrong. I am claiming that the success of it is by no means proven , and it has many problems of its own (including those you list as problems for alternative theories).
[1] https://tritonstation.com/2024/06/18/rotation-curves-still-f...