Assume we have a quantum state $$ |\psi\rangle = \sum_{j=0}^{2^n -1} \psi_j |j\rangle $$ where $\psi_j \in \mathbb R \ \forall j$.
My knowledge of the HKP shadow tomography protocol is limited, but it seems like given a set of $M$ observables, one can estimate their expectation values with $O(\log M)$ measurements. If that's the case, if we let our set of observables be $\{|0\rangle \langle 0|, |1\rangle \langle 1|, \dots, |2^n-1\rangle \langle 2^n -1 |\}$, wouldn't we be able to do full tomography on $|\psi\rangle$ using only $O(n)$ measurements? I'm certain I'm missing something here but I'm not sure what. Is it the fact that the amplitudes can be exponentially small and the accuracy becomes the limiting factor?