My two-qubit test rig on the bench keeps losing coherence faster than the readout indicates, even after swapping the cables and tightening the fridge connections. Last night I tried a basic repetition code to catch bit flips, but the errors piled up before the correction step could run. It leaves me wondering how to scale anything without the system falling apart from tiny control drifts or material defects that still slip through screening. What specific fixes have people tried on similar small rigs to extend coherence long enough for even one correction cycle?