Yes! This is exciting to see. Erin Catto is such a cool hacker. Thank you, Erin, for sharing your code with the open source community.
There wasn't anything about determinism in the announcement, but I'd really love to see some more about that, too. Trying to use Unity's built-in physics to make a networked billiards game is quite troubling, when none of the clients can happily agree on what happened.
I was looking for the same thing. There is a replay mechanism, so it seems to be deterministic. But with floating point physics, not across platforms. Though -ffast-math is unsupported according to the documentation, so maybe it is intended to be deterministic across platforms? https://box2d.org/documentation3d/recording.html
PhysX itself (assuming Unity still uses PhysX, but any major physics engine will be the same) is unlikely to be your problem, although it is possible that Unity compiles it with undesirable optimisations, or some aggressive multithreading strategy, etc.
In any case, it's quite likely that the physics simulation is not being initialised with identical state on each instance, so the physics engine -- even if it claims to be deterministic -- cannot be expected to arrive at the same solution across instances.
From the scripting layer atop Unity (et al.) you are essentially out of luck. You don't have the requisite control/access to guarantee the simulation input is consistent.
To expand on the "quite likely" variance in the simulation input, there's an unending list of possibilities, so just a couple of points: Unity is really really big and complicated.
The physics engine in Unity is not primarily there to be a FEATURE: physics engine. It's a general purpose collision system, and it's fundamental to Unity's whole world/scene/components. It's always running. Stuff that has nothing to do with what you see as your simulation can affect your simulation.
My unopinionated advice is: read Glenn Fiedler's classic networked physics articles[0].
My opinionated advice is: you're making a game where the physics is the whole game. It's also a very nicely contained problem space. Write your own physics simulation. It's gameplay code. Also, typical rigid body physics engines certainly can do a billiards sim, but they aren't actually that good at it.
There wasn't anything about determinism in the announcement, but I'd really love to see some more about that, too. Trying to use Unity's built-in physics to make a networked billiards game is quite troubling, when none of the clients can happily agree on what happened.