Researchers used a differential‑testing tool, QD IFF, to run thousands of altered versions of six benchmark quantum algorithms on three open‑source software stacks—Qiskit, Cirq, and Pyquil. The test framework creates semantically equivalent circuit variants, filters out circuits unlikely to surface errors, and compares quantum outcomes with statistical tests.

The tool generated 730 program variants and 14,799 corresponding circuits. After filtering, only 19–42 % of the circuits were executed on real machines or simulators, cutting hardware or noisy‑simulator calls by 66 % and trimming the 17‑day overall testing time.

Running the filtered circuits on IBM’s “ibmq_yorktown” device uncovered 33 divergent cases. Twenty‑five of these diverged because two‑qubit gates on weaker qubit couplings produced errors, while four were attributed to qubit dephasing or decoherence during idle periods.

These findings show that even well‑used quantum programming libraries can hide subtle reliability problems that are hard to detect without systematic testing. The study underscores the need for robust testing workflows in quantum software development.