High-performance simulation-decoding framework for erasure-aware quantum error correction

We present a fast, extensible, end-to-end framework for simulating and decoding erasure-aware surface codes. Erasure-aware modalities have seen a surge of recent interest; despite many simulation efforts (Wu 2022; Sahay 2023; Kubica 2023; Kang 2023; Gu 2024; Baranes 2025), the field lacks a reference implementation and performance baseline. Built around dual-rail cavity qubits (DRCQ) yet portable across architectures, our approach augments stabilizer simulation with per-shot leakage dynamics: gate skipping on leaked qubits, leakage propagation, conditional Pauli errors, and imperfect or deferred erasure checks. These elements yield a detector error model (DEM) with static and per-shot erasure-aware terms, enabling dynamic reweighting in a matching decoder. Our implementation builds on and extends Stim (Gidney 2021) and PyMatching (Higgott 2025), minimizes cross-shot redundancy, and is written in Rust/C++. In memory benchmarks under realistic DRCQ noise, it achieves ~100x higher throughput than prior erasure-aware workflows. Because its primitives—leakage-driven sampling, DEM generation, and erasure-informed reweighting—are hardware-agnostic, the toolchain applies to Rydberg arrays and other erasure-aware platforms via user-defined leakage models and erasure checks, enabling rapid exploration of codes, schedules, and device trade-offs at the circuit level.