usable output, drifting qubits
96 qubits, ibm_kingston
Keep the register usable inside the shot budget.
Maintain measured qubit responses without a width-dependent calibration sweep.
Contract entry at acquisition four · 48 controlled channels · real QPU
residual vs measured drift
two-cycle lock, flat for nine hours
maintenance downtime
50 / 50 cycles through the public cloud
Connect measure(x) to the per-qubit response from each parallel shot batch.
TrueLoop consumes measured classical statistics; it does not read a quantum state or provider-private calibration registers.
Open the quickstartCan one shot batch resolve every relevant qubit response before the operating point moves?
Check compatibilityWhere it applies
Calibration holdState preparationReadout stabilizationQKD linksQRNG operating pointsResponse-profile matching
What this evidence shows. And what it does not.
Native-drift maintenance on 96 qubits of ibm_kingston: SUPPORTED on the registered drifting-subset branch (34 qubits above threshold, +25.7 points usable), residual 0.27× measured drift, zero downtime; the aggregate stayed inside the loose fresh-floor tolerance, and the subset hypothesis is the claim. Earlier campaigns: commanded-disturbance loops at four and 48 channels. Width scaling beyond hardware remains controlled simulation.