QuBridge: Layer-wise Fidelity Decomposition in Quantum Computation Pipeline

Abstract

Running a quantum circuit on current hardware involves a sequence of engineering decisions, each with tunable parameters and distinct error characteristics. Existing tools optimize each decision in isolation, leaving practitioners unable to determine how much each decision contributes to final output quality. We present QuBridge, a pipeline analysis tool that decomposes quantum computation into three decision layers and measures each layer's fidelity contribution through progressive ablation and isolation experiments. Applied to quantum teleportation under IBM-calibrated noise models, the framework surfaces three phenomena that end-to-end measurement obscures. Qubit selection narrows the worst-case fidelity band from 11.8% to under 2% with downstream layers held fixed, without changing the peak. Per-gate pulse-shape assignment adds a +0.9% residual gain whose attributed magnitude depends on upstream layout. Error-detection encoding is not uniformly advantageous, and its conditional benefit emerges for input states whose dominant error channel is detectable by the chosen code. QuBridge operates on cached calibration data without requiring live hardware access.