Consider a theory with a $\mathrm U(1)$ symmetry, i.e., such that there exists a unitary operator $U\equiv\mathrm e^{iQ}$ that commutes with the $S$ matrix (or the Hamiltonian). The hermitian operator $Q$ represents a conserved charge, which we may refer to as the electric charge (or baryonic charge, etc.).
The states of the theory are classified according to unitary representations of the symmetry group, which in this case contains a $\mathrm U(1)$ factor. Now, the unitary representations of this group are of the form $z\mapsto z^n$ with $z\in\mathbb C-\{0\}$ and $n\in\mathbb Z$, which means that states are labelled according to \begin{equation} Q|n,\dots\rangle=n|n,\dots\rangle \end{equation} where "$\dots$" refers to other labels. From this I would conclude that electric charge, or any other $\mathrm U(1)$ charge, is always quantised. There exists a minimal charge, say $q$, such that the charge of any other state is $nq$ for some $n\in\mathbb Z$. This seems to be in agreement with what we observe experimentally.
Now comes my question: I would have expected that we should allow for projective representations rather than regular ones. This means that we may now allow $z\mapsto z^n$ with $n\in\mathbb R$, i.e., the quantum number is no longer quantised. Any charge should be observed rather than only those a scalar multiple of some minimal one. This does not seem to agree with what we observe experimentally. Why is this? Why must we only consider regular representations rather than projective ones? Should or should not $\mathrm U(1)$ charges be quantised?