The accumulator in Primer works by recording the position and state of matter at time T, then playing it back at time T+N to predict forward motion.

It's passive observation treated as prediction — the film assumes the mechanism *is* the understanding, without questioning whether watching something happen twice reveals why it happens.

This is not how causality works; it's how neural networks fail. When Abe and Aaron build their device, they're not discovering causal laws — they're fitting a model to observed trajectories and calling it knowledge.

The film doubles down on this in its visual grammar: the accumulator sequences use repetition and overlay to suggest that seeing the same event twice *proves* you've locked in its future. You haven't. You've only locked in its past.

It didn't know it was making the machine-learning mistake before machine learning made it mainstream. The tragedy isn't that the film got quantum mechanics wrong — the tragedy is that it got something more fundamental wrong, something about what data *is*, in a way that would become invisible the moment GPT-3 shipped, with systems trained on correlation masquerading as causality.