It is known that testing whether a list of $n$ arbitrary real numbers is $\varepsilon$-close of being sorted (in Hamming distance) has query complexity $\Theta(\log n)$ [1]. It is also easy to show that if the function is Boolean-valued, then this becomes $\Theta(1/\varepsilon)$.
Recent work of Berman et al. [2] considers the same problem for sequences with elements in $[0,1]$, but with another distance (namely, a (normalized) $L_p$ distance instead of Hamming — the normalization being done to ensure the distances lie indeed in $[0,1]$). They show the query complexity then becomes $O(1/\varepsilon^p)$, for $p\in\{1,2\}$.
Now, does anyone know if any of these two variants has been considered (either Hamming or non-normalized $L_1$), but with the further restriction that the sequence has bounded $L_1$ norm? That is, decide (with high probability) whether a sequence of non-negative numbers $a_1,\dots, a_n$ with $\sum_{k=1}^{n} a_k = 1$ is (a) sorted or (b) $\varepsilon$-far from sorted?
I tried to find references, but so far didn't find any -- and want to make sure there is no prior work I'm unaware of before diving further into this (I am in particular interested in adaptive algorithms, as I (think I) have a non-adaptive tight bound for this question).
