You could order your events A B C and say this denotes the direction of time, but there is no actual time flowing or moving in any direction. Your light beam has to move from A to B to be able to move from B to C. That's what people say, but look closely. This principle states that time has a direction.

There is no such thing as negative motion. When the light reaches B, that's an event, and there's nothing that can undo the motion of that light from A to B. IMHO you'd be better off thinking about a light beam moving from A to B to C. That principle is a pat statement that doesn't actually explain anything. I got to the point where in order to avoid paradoxes one can introduce the causality principle A past light cone models light coming at you from all directions. The future light cone models an expanding sphere of light. You cannot point up to the clear night sky and say, 'Look, there's a light cone'. But I will say this: a light cone is an abstract thing. Relativity is just about the best-tested theory we've got. I was reading about the light cone in relativity. IMHO it's important to look hard at the ontology of what's actually there and take care to distinguish between reality and abstraction. This is known as Loschmidt's paradox, which itself has many possible resolutions. Of course, this doesn't address the other side of this issue, which is why there would be a causality principle at all, given that the microscopic laws of physics are reversible in time. Jaynes doesn't mention relativity explicitly, but if we take his view then the causality principle can be seen as a common assumption in both relativity and thermodynamics.

(The only way we can affect the final conditions of an experiment is via the initial conditions and the boundary conditions.) Together with Liouville's theorem this is enough to derive the second law. (See sections V and VI for the discussion, which I think can be read in isolation from the rest of the paper.) Here he derives the second law from the empirical fact that we as scientists and engineers are able to manipulate the initial conditions of an experiment, but we can't directly manipulate the final conditions. There are several possible approaches to this question, but I've always been a fan of the one taken by Edwin Jaynes in his 1965 paper Gibbs vs Boltzmann Entropies.

Causality principle