My question is a bit on the soft side, and basically asks if there is something between error mitigation and full error correction?
On the one hand, as far as I understand, trying to increase the quantum volume by error mitigation hits a wall based on the decoherence rates and gate errors. For instance, the recent 127-qubit simulation on IBM's chip was limited by what error mitigation can do. Adding more qubits wouldn't really help, because it was really the depth that was a key limitation.
On the other hand, having enough noisy qubits can get you to the full error correction, but the overhead is typically orders of magnitude.
Is there something in between? For example, by using codes with small distances, can we trade a dozen of physical qubits for a single composite one, which is still error-prone but less so than the original? Then we could expect a 1000-qubit chip to do things that a 100-qubit one could not. The way of packaging qubits need not even be an error-correcting code (e.g. it doesn't have to eliminate all errors up to a certain weight) but could be something like an error-suppression code, defined simply by the fact that a composite qubit is better than any individual one.
If these things have already been discussed somewhere, I'm happy to take references.