The evolution of gate set tomography

Gate set tomography or GST, introduced 12 years ago as a calibration-free protocol for characterizing single-qubit gates, has grown into one of the mainstays of QCVV. Since 2013, it has evolved into a (perhaps surprising) array of diverse niches. Nonlinear "long-circuit" GST unlocked almost unlimited accuracy with reasonable experimental cost. The pyGSTi open-source software made reliable GST accessible to any experimentalist. Fiducial pair reduction transformed 2-qubit GST from a hero experiment to a feasible, routine technique. Reduced models offer the potential to make GST feasible on 3-4 qubit operations. Variants like randomized linear GST, along with closely related protocols like ACES and Pauli noise learning for gate sets, enriched the theoretical basis of GST-adjacent protocols. GST variants have characterized crosstalk and mid-circuit measurements, tracked drifting parameters, and fit the parameters of low-rank and non-Markovian models. I'll put this history in context, and then give a preview of efforts at Sandia's Quantum Performance Lab to extend GST to characterize leakage, non-Markovianity, and many-qubit quantum computers.