Deterministic Benchmarking of CZ gates

2 qubit gates form the cornerstone of quantum error correction and quantum algorithms, but are also one of the biggest, potentially fixable, components in error budgets in current systems. Current benchmarking protocols provide information about gate performance in an "extreme" sense: Randomized Benchmarking condenses the gate performance into a single infidelity number, while Gate Set Tomography unpacks the same performance into a metric composed of exponential numbers. We propose a CZ gate benchmarking technique, which extends previous work on 1 qubit gate deterministic benchmarking (DB), and instead characterizes the performance in terms of coherent and incoherent error mechanisms. We experimentally demonstrate DB for CZ gates using superconducting transmon qubits and support these results with simple analytical models and master equation simulations. Our findings also point mitigation strategies for CZ gates, specifically for the coherent error mechanisms.