If we limit to just QStabilizer attachments to QBdt, as we do not PushStateVector() or PopStateVector() to multiply-out amplitudes, we also do not normalize stabilizer phase per usual QBdt rules. (If we limit test_mirror_quantum_volume to just Clifford and Pauli gates, this now seems to have a 100% success rate.)
If we limit to just
QStabilizerattachments toQBdt, as we do notPushStateVector()orPopStateVector()to multiply-out amplitudes, we also do not normalize stabilizer phase per usualQBdtrules. (If we limittest_mirror_quantum_volumeto just Clifford and Pauli gates, this now seems to have a 100% success rate.)