> The models for this are IIRC trigonometric polynomials of fairly low order, so even if we could model the unpredictability perfectly, truncation error would limit our ability to distribute the model at super high accuracy. The existing models are built in to, eg, satellites, so you can't just make them arbitrarily complex.

What are the satellites doing with the models? They're not deciding leap seconds on their own, I hope. So I don't see why the leap second decision would be locked to low accuracy.

Also I would expect doubling the precision to give you a 3-4x slowdown on the math or adding orders to have less effect, and the amount of available computation spent on those models to be like a tenth of a percent at most, so the extra cycles wouldn't be an issue. What am I missing here?

To me it seems like unpredictability is the only real issue.