Randomized Benchmarking
The standard protocol behind nearly every gate-fidelity number: run long random gate sequences that should do nothing, and watch how fast the success rate decays.
Randomized benchmarking is the standard protocol behind nearly every gate-fidelity number quoted in the industry. Run long random sequences of gates that should, in total, do nothing; measure how often the qubit actually returns home; watch the success rate decay with sequence length. The decay rate yields an average error per gate that's conveniently immune to state-preparation and readout mistakes.
Its virtue and its vice are the same: it produces one clean, comparable number by averaging away the details. A gate can pass randomized benchmarking handsomely while harboring coherent errors that compound viciously in structured algorithms. When a company quotes '99.9% fidelity,' the right follow-ups are: measured how, on how many qubits at once, and does it hold when the neighbors are busy?