Two different mechanisms, arriving at the same place from opposite directions. The items at the start get rehearsed while the list is still short, so they have time to be written down properly — that's primacy. The items at the end are still ringing when you're asked, because nothing has come along to push them out yet — that's recency. The items in the middle got neither. They arrived too late to rehearse and too early to still be echoing.
You can break either half on purpose, which is how the two were told apart in the first place. Delay the question by thirty seconds of arithmetic and recency vanishes while primacy survives. Speed the list up so nobody can rehearse and primacy vanishes while recency survives. Two mechanisms, one curve.
Eight words, one at a time, then type back whatever you have. Four lists, because one is nothing — and the curve only appears once there are enough of them.
Four lists is not enough
With four lists each position has been tested four times, so every point on that curve can only be 0, 25, 50, 75 or 100 per cent. That is a very coarse instrument, and a shape that looks like a U may be four coin flips that landed conveniently. Run it again and the page adds your new lists to the old ones rather than starting over — the curve settles somewhere around a dozen lists, and until then it is a suggestion.
This is the honest version of a complaint I have about the whole genre. Laws of this kind get quoted from a diagram in a slide deck, and the diagram is an average over hundreds of people. Yours will not look like it. Mine doesn't.
What it costs you
The middle of a list is the cheapest place to put something and the most expensive place to have put it. If a menu has nine items, four of them are load-bearing and five are decoration you paid to build, maintain and translate.
The first and last steps are the ones described to a colleague. Whatever happens in the middle of your onboarding is not what people will say your product is like. Two moments carry the account of the whole thing.
Recency dies faster than primacy. If the decision happens immediately — a menu open right now — the end of the list is strong. If it happens tomorrow, only the beginning survives. Which end you optimise for depends entirely on when you think the person acts, and almost nobody asks that question before ordering a list.
Alphabetical order is a decision not to decide. It is defensible, it is neutral, and it puts your most important item wherever its first letter happens to fall.
I once ordered a feature list by how hard each one had been to build. Nobody has ever remembered the middle of it, including me, and I am the one who fought for those.