Quantum circuits for search still hit the same computability wall as any Turing machine.
I put a few qubits into superposition, applied an oracle that marks the target state, then measured after the diffusion step. The probability boost showed up exactly as the math predicted for n=3, but nothing about the output bitstring was unreachable by looping through the same small space classically. Noise on the actual backend just made the counts noisier without creating new possibilities.