Qiskit Quantum Backend Selector

backend-orchestration
runtime-tooling
noise-modeling
Queue- and noise-aware backend selection and routing for IBM Quantum backends, verified against real hardware
Published

September 3, 2026

STATUS · Qiskit Ecosystem submission in review · PR #1363 (draft)

What it does

QiskitRuntimeService.least_busy() only looks at queue depth. This package scores candidate backends per-circuit instead — whether the circuit actually fits a given device, plus an estimate of expected fidelity after transpilation, pulled from live calibration data — then can route job submission with automatic ISA transpilation and failover if a backend errors out mid-attempt.

Three pluggable scoring strategies (QueueOnlyScoring, NoiseAwareScoring, HybridScoring), a ranking selector, and a router that retries the next-best candidate on submission failure.

What’s actually been checked, and what hasn’t

The fidelity estimate is regression-tested against Qiskit Aer noise-model simulation (~0.02 agreement across bell-pair and GHZ test cases) — but that’s a self-consistency check of the analytic math against a simulation built from the same calibration data, not independent validation against real device behavior (crosstalk, drift, and correlated errors aren’t in either the estimate or the check).

The routing and submission path, by contrast, has been run against a real QiskitRuntimeService twice. The first attempt caught a genuine bug: IBM Runtime requires ISA-conformant circuits since March 2024, and the router was handing the raw circuit to the submission callback — ibm_fez rejected it outright. Fixed, then reran: a real job completed successfully on ibm_fez in 3 seconds.