An empirical sweep of SAT and MaxSAT based exact distance methods finds that parity reasoning tricks from classical LDPC don't transfer to QLDPC, and that the choice of MaxSAT architecture matters more than the SAT versus MaxSAT split itself.
QLDPC codes are rich in XOR structure, so for years the natural assumption was that parity-aware reasoning would dominate exact-distance computation. A new systematic benchmark, "SAT, MaxSAT, and SMT for QLDPC Distance Computation: A Large-Scale Empirical Study", finds that this intuition is misallocated. The thing that actually governs scalability is not whether a solver is XOR-aware. It is how the solver handles cardinality constraints, how tightly it can tighten optimization bounds, and what MaxSAT architecture sits under the hood.
The paper, posted to arXiv on 2026-05-29 by Yu-Fang Chen and colleagues, runs a large empirical sweep across representative QLDPC codes and compares SAT- and MaxSAT-based formulations against several established exact-distance approaches. It frames its contribution as a refinement of prevailing intuitions rather than a verdict on any one tool. Three findings, in particular, deserve the attention of anyone building QLDPC validation pipelines.
First, despite the abundance of XOR structure in QLDPC parity checks, XOR-aware reasoning does not provide a systematic advantage across the benchmark suite. Practical scalability tracks cardinality-constraint handling and optimization-bound strength, not parity cleverness. For a community that has invested considerable effort in XOR-specialized propagators and encodings, that is a reorientation, not a refutation: on individual instances XOR techniques can still help, but they are not the dominant lever.
Second, Brouwer-Zimmermann-style search, the classical-LDPC benchmark paradigm for exact distance in sparse codes, no longer holds its traditional scalability edge in the QLDPC setting. The classical playbook was tuned for codes whose structure looked very different from typical QLDPC constructions, and that advantage does not transfer cleanly. Read narrowly, this is about scalability parity rather than failure; read operationally, it means code designers should not treat Brouwer-Zimmermann search as a free baseline.
Third, and most actionable, the choice of MaxSAT solver architecture matters far more than the SAT-versus-MaxSAT framing suggests. Branch-and-bound MaxSAT substantially outperforms unsat-core-based MaxSAT on the paper's challenging benchmarks. For practitioners choosing a back end, that is the kind of finding that reorders a procurement decision.
Taken together, the three findings point to a constructive reallocation of effort. QLDPC tooling work, the authors argue, should invest less in parity-reasoning specialization and more in cardinality-constraint handling, bound-tightening techniques, and MaxSAT solver architecture. That is a research-program pointer rather than a winner announcement, and the paper's scope is deliberately bounded to its benchmark suite.
A few caveats are worth carrying through. The paper is a v1 arXiv preprint submitted 2026-05-29, roughly two weeks old as of this writing, with no listed peer-reviewed venue on the abs page. The arXiv record shows no subsequent versions as of the current check. The "prevailing intuitions" it interrogates reflect a 2024–2025 reading of the field, and QLDPC design is moving quickly; any post-submission arXiv activity that extends or contests the benchmark would be worth a short look before treating the results as stable guidance. The authors are also the source of the headline counter-intuitions, so independent replication or an outside quote on practical impact for code designers would strengthen the story, though the underlying claims remain falsifiable and concrete.
What to watch next: whether the same benchmark suite, re-run against a refreshed set of QLDPC families, still singles out cardinality handling and MaxSAT architecture as the binding constraints, and whether the broader community revises its toolchain defaults accordingly. For now, the working conclusion is precise. Parity tricks are not dead, but they are no longer the right place to spend the next unit of effort.