In evaluating what is needed for real libertarian free will to be valid, I have identified two things:
The physical world must be causally open to conscious causation acting on it.
Causation must include more options than "caused vs random".
For point 1, There have been multiple attempts to use the indeterminacy to QM to leverage up to an indeterminacy in the universe. There was another recent question asking about these efforts from the last several decades, which referenced Koch and Hepp's paper Is quantum mechanics relevant/irrelevant to explain conscious processes?. I agree with Koch and Hepp that the approaches they evaluated look to be unsuccessful.
But I suggested in my answer that the leveraging up of QM indeterminacy to macro scale effects, such as all of life depending on photosynthesis, or the catastrophic effects of a single gamma ray on an organism's survival if the ray triggers cancer, provides an alternative path for this indeterminacy to be effectuated. Koch and Hepp did not evaluate chaotic leveraging of QM indeterminacy in their paper, and I believe if they had, they would have been forced to a different conclusion. This leveraging is mathematically and empirically verified.
For point 2, this is where your speculation about probability summation may be needed. I am not a logician, nor probability theorist, so I am not sure how these matrix into the problem.
The basic solution to point 2 is Agency Causation. That causation has three options, not two, and one of the three is "caused by an agent's will". I found a good recent book that articulates a version of agency causation, by Helen Steward: https://global.oup.com/academic/product/a-metaphysics-for-freedom-9780198706465?cc=es&lang=en&#. I did not find any discussion of probability summing in her work, but she is not the only agent causation theorist, so you may want to dig into the writings in this community.
Note, I did not find Steward's book convincing in itself. And this is related to a logic issue, which your probability summing may apply to.
I had several concerns with Steward's thinking.
A) she stayed with a basic physicalist model, while for an agent to even exist, she needed to at least accept Popperian emergent dualism with a world 2, for there to be a place for an agent to BE to influence matter. She also had apparently not explored the problems physicalism has with abstract objects, which include information. Physics itself has basically become a world 1/3 dualism, as information is treated as real, but it has no mass nor energy. If she had adopted Popperian 3 worlds, which is a naturalist triplism, then I would have found her model to be more self-coherent. This is not a fatal problem, as she already accepted strong emergence for consciousness, so her approach can be recast in triplist terms that explicitly admit to the causal openness of physics.
B) The bigger concern I had with her approach is that she really does not deal with "what causes agents to choose" effectively. This is the problem that recursion answers face in the Munchausen Trilemma -- "and then what is the justification for THAT answer/reason/cause". The general approach of science is to pursue the 3rd leg -- and to accept as an interim working solution a regularity we discover, while then trying to find a reason why that regularity holds. But agent causation seems to have to stop at the first leg -- "this is just the way the world works and we don't have an explanation".
One of the reasons I don't bother digging into logic is that naturalism and logic themselves have trouble with the trilemma. My empirical/pragmatic approach, and all of science, is only justifiable per leg 2, and that is a recognized fallacy of circularity. Pluralism of logic leaves us with an infinity of disparate logic evaluations, and no method to evaluate these diverse evaluations. Here is my summary of the Trilemma, and the problems it provides for rationality: Is the Münchhausen trilemma really a trilemma?
The question I have for you, is not whether free will can be resolved using nonunitary probability, but can nonunitary probability rescue logic from the trilemma? I have serious doubts it would be useful at all, and if you can't resolve the trilemma, then you can't solve my concern B) for agent causation.