In the 2016 Science paper "Realization of a scalable Shor algorithm" , the authors factor 15 with only 5 qubits, which is fewer than the 8 qubits "required" according to Table 1 of  and Table 5 of . The 8-qubit requirement comes from the end of  which states that the number of qubits needed for factoring an $n$-bit number is $1.5n+2$ which for 15 is $1.5\cdot 4 + 2=8$.
The paper that uses only 5 qubits states that their algorithm "replaces a QFT acting on M qubits with a semiclassical QFT acting repeatedly on a single qubit", but the consequences of this on the complexity of the algorithm was never mentioned in the paper.
Now there has been harsh criticism of the paper claiming to factor 15 in a "scalable" way, as they say in Section 2 that the complexity argument for Shor's algorithm no longer holds. However, this criticism has not been corroborated anywhere, and the Science paper keeps getting celebrated as a "scalable" version of Shor's algorithm. What is the complexity of the "scalable" Shor algorithm?