I got 39. I found the test interesting in terms of sorting algorithms. Typically when we think of sorting stuff, the comparison function is deterministic and cheap.
In this case, it gets increasingly hard and even erratic to tell adjacent colors apart; essentially the comparison function is increasingly expensive and non-deterministic.
I wonder if that has any consequences for the "optimal" strategy. In my case -- and I suspect most people did it the same way --, I used a kind of two-way insertion sort: for each field, I first decided if it should move left or right, then moved it as far as it "made sense". So I ended up with a list that got sorted from the outside towards the inside. In the end I had a few passes of "bubble sort", checking adjacent pairs for correctness. This was when it got expensive and erratic. :)
It would be a fun experiment to have an algorithm sort the list (based on whatever sorting algorithm) and simply use the human as a comparison robot: display two hues at the same time and let the human decide which of the two is "smaller". I'm sure you could easily measure and even predict the increase in "expense" (i.e. response time) and the increase in errors and ambiguity for close hues.
You could also do stuff like force the user to decide within a very short time, to determine if there's actually an increase in accuracy when staring at a given pair for 10 seconds vs a 500ms "hunch" decision (or heck, 10 500ms hunches, still a two-fold improvement).
In this case, it gets increasingly hard and even erratic to tell adjacent colors apart; essentially the comparison function is increasingly expensive and non-deterministic.
I wonder if that has any consequences for the "optimal" strategy. In my case -- and I suspect most people did it the same way --, I used a kind of two-way insertion sort: for each field, I first decided if it should move left or right, then moved it as far as it "made sense". So I ended up with a list that got sorted from the outside towards the inside. In the end I had a few passes of "bubble sort", checking adjacent pairs for correctness. This was when it got expensive and erratic. :)
It would be a fun experiment to have an algorithm sort the list (based on whatever sorting algorithm) and simply use the human as a comparison robot: display two hues at the same time and let the human decide which of the two is "smaller". I'm sure you could easily measure and even predict the increase in "expense" (i.e. response time) and the increase in errors and ambiguity for close hues.
You could also do stuff like force the user to decide within a very short time, to determine if there's actually an increase in accuracy when staring at a given pair for 10 seconds vs a 500ms "hunch" decision (or heck, 10 500ms hunches, still a two-fold improvement).