Total: 1
We study a two-qubit sensing protocol in which only a helper qubit is read out. Calibrated loading and unloading encode a transverse signal in the helper's phase, keep leakage out of a selected dressed-state band small, and return the sensor close to its idle ground state. We derive an information bound for a single sensor whose excitation stays below a prescribed budget, and simulate repeated operation with an uncertain coupling and finite-duration resets in a driven open-system model. At the tested settings, finite measurement records support estimation with infrequent resets. A helper flip at zero coupling, combined with a reversal of the coupling sign, suppresses a static helper offset shared across a record while retaining the signal, at a cost of about 27% of the no-offset information of a plain sequence of equal duration. Under slowly varying noise, averaging each cycle's likelihood over the correct noise law removes most of the added bias, and a wrong noise law can worsen the estimate. A selected directly read sensor with ideal controls and repeated nondemolition verification has the higher information rate. Any practical benefit of avoiding direct readout therefore requires a device-specific disturbance model.