glean

Documentation

Guides & tutorials

Five hands-on walkthroughs, in order — from a zero-dependency toric-code decode to circuit-level qLDPC decoding, picking the right decoder, reaching rare logical-error rates, and plugging into Stim's collection framework. Each snippet runs against the built module.

Build from source first: maturin develop --release

Before you start

  • Build the Rust core into your active environment with maturin develop --release, then check import glean works.
  • The first guide needs nothing else. From the second onward, install the optional extras you use: pip install stim (circuits + detector error models), numpy (array ergonomics + the estimators), and sinter (the last guide).
  • New to the API surface? Keep the API reference open alongside these guides.
  1. 1

    Quick start

    Build a toric code, inject an error, decode it, and measure a logical error rate — with no external dependencies.

    5 min →
  2. 2

    Circuit-level decoding from a Stim DEM

    The real workflow: ingest a flattened Stim detector error model and decode batches of syndromes entirely in Rust.

    10 min →
  3. 3

    Choosing a decoder

    BP+OSD-CS, relay-BP, or BP+OTF — what each of the three paradigms is good at, and how to swap between them on one DEM.

    8 min →
  4. 4

    Rare-event estimation

    Reach 10⁻⁷ logical error rates without 10⁹ shots, using the decoder-agnostic glean.dss and glean.splitting estimators.

    12 min →
  5. 5

    sinter integration

    Plug Glean into Stim’s sinter.collect(...) pipeline as a drop-in, picklable decoder.

    6 min →