I’ve had a think about this and here is my theory. Sometimes with these things there’s some ambiguity with the terminology. In this case the node is described as working “at a constant rate.” It’s not clear just from reading what exactly is “constant.” Personally, I would intuitively expect it would take a constant amount of time to move from a starting value to the target value, similar to the function of the “non-constant” version, just without the smoothing. I may be wrong but I feel like most people would expect it to behave similarly to the other, given similar naming and identical inputs.
What I think is actually happening is that the numerical change over time is what is being kept constant, even if the input changes. What would this be for? For positions. Otherwise known as locations. If, for some reason, you wanted to interpolate between one location and another and have your thing move through space at a “constant” rate, that’s probably what this is. Think of moving between two points on a map. If the points are far away, it will take a long time.
Here I am trying to interpolate between velocities and rotations (technically normalized vectors) because it is labeled for vectors! What I’m doing doesn’t even make sense! But there’s nothing at all unreasonable about the approach at the surface level, based on the information presented.
They really ought to rename this, or update the description at the very least.
Edit: if it matters to anyone, I bet you could multiply the value going into Interp Speed by the difference (distance) between the two input vectors to normalize the interpolation time and make it function the way I thought it would. They also ought to have a proper node that does that.