I built three tools for Quantinuum's guppy stack. Along the way I found six real bugs.

A developer created three tools for Quantinuum's Guppy stack, a quantum programming ecosystem. While building these tools, the developer discovered six genuine bugs. The tools were designed to enhance the Guppy stack, focusing on correctness rather than just demonstrating functionality. The first tool, Qshelf, is a tested package registry for quantum algorithm implementations such as QFT, Grover, QAOA, and VQE-H2.

Each tool was verified against independent mathematical references to ensure accuracy. The second tool, Estimand, serves as a fault-tolerant resource estimator for Guppy/HUGR programs, providing estimates on physical qubit count, runtime, and error probability under a surface-code scheme. The third tool, Qmatchpoint, integrates PyMatching, a peer-reviewed decoder, with syndrome bits produced by a Guppy QEC circuit.

The developer found four defects in Guppylang during the creation of Qshelf. These issues include incorrect units from IQFT compiled individually versus combined with QFT, a wrong unitary from multi-controlled Z, a rejected generic array-length type, and a rejected numpy.ndarray closure. The developer also encountered two significant discrepancies while verifying the estimation formulas used in Estimand.

The first discrepancy involved a missing additive constant in the CompactDataBlock's tile-count formula, as cited in the paper by Beverland et al. (2022). The second discrepancy arose from a magic-state factory error model in Estimand, which inadvertently used Beverland's threshold constant instead of Litinski's, despite the developers' intention to adhere faithfully to Litinski's original paper.

These discoveries highlight the importance of rigorous verification in software development, particularly in the rapidly evolving field of quantum computing.