v1.2.0 — Stable Release

v1.2.0 is the first stable release of the new hardware integration flow.

• Moved hardware integration to top-level: hardware_integration/xanadu/
• Added one-command real-data workflow: xandau_hardware_data.sh
• Supports Xanadu dataset conversion (Aurora, QCA, GKP, legacy job JSON) to decoder_io NDJSON
• Keeps outputs centralized in examples/results/hardware_integration/
• Includes replay workflow and updated documentation across README and docs

v1.2.0-rc.2 – Streaming Xanadu Data Integration RC2

This release candidate focuses on Xanadu hardware-data integration and large-dataset conversion reliability.

Highlights

• Added multi-format Xanadu data conversion into decoder_io NDJSON:
  • legacy job JSON (output/samples)
  • Aurora decoder-demo switch-setting batches
  • QCA/Borealis shot matrices (.json, .npy, .npz)
  • count-compressed outcome tables (GKP-style exports)
• Added C++ replay workflow integration for converted NDJSON requests.
• Added streaming/chunked conversion controls for heavy datasets:
  • --stream
  • --shot-start
  • --max-shots
  • --append-out
  • --progress-every
• Added runnable fixtures/scripts for Aurora/QCA/GKP conversion and replay validation.

Included Changes Since v1.1.4

• feat(hardware): add decoder_io NDJSON replay CLI and Xanadu integration example (431b0c9)
• feat(hardware): add streaming Xanadu dataset integration (a544e2b)

Comparison:
https://github.com/DennisWayo/lidmas_cpp/compare/v1.1.4...v1.2.0-rc.2

Validation Performed

• Python compile check:
  • python3 -m py_compile examples/hardware_integration/convert_xanadu_job_to_decoder_io.py
• Fixture conversions:
  • bash examples/hardware_integration/run.sh
  • bash examples/hardware_integration/run_public_datasets.sh
• Replay checks:
  • bash examples/hardware_integration/replay.sh examples/results/hardware_integration/decoder_requests_aurora.ndjson
  • bash examples/hardware_integration/replay.sh examples/results/hardware_integration/decoder_requests_qca.ndjson
  • bash examples/hardware_integration/replay.sh examples/results/hardware_integration/decoder_requests_gkp.ndjson
• Result: replay completed with errors=0 on fixture datasets.

Notes

• .npy/.npz conversion paths require NumPy (pip install numpy).
• This is an RC intended for real-data validation at scale before final v1.2.0.
• No breaking changes are expected for existing workflows.

Upgrade / Usage

• Use this RC tag for validation runs:
  • git checkout v1.2.0-rc.2
• For large QCA/Aurora inputs, prefer streaming/chunking:
  • --stream --shot-start <N> --max-shots <K> [--append-out]

v1.1.0 — Native GKP Pipeline + CUDA Sampling

• Native GKP mode with digitizer, gate/idle/meas errors, and loss map support.
• CUDA-accelerated Pauli surface_threshold sampling (CPU decoders unchanged).
• GPU self‑test and CPU vs GPU benchmark presets.
• Paper workflows split into paper_01 (frozen) and paper_02 (current).
• Surface demo improvements and extended CLI controls

v1.0.0 – Hybrid CV + Discrete QEC + Example Suite

Major Features

• Hybrid Continuous-Variable (Gaussian σ-noise) and Discrete Pauli noise support
• Surface-code threshold estimation
• Finite-size scaling collapse fitting
• Neural-assisted MWPM decoder integration
• Deterministic reproducibility with fixed seed
• Examples

v0.9 — Stable scaling & threshold extraction release

This release marks the first stable implementation of surface-code scaling and logical error rate (LER) threshold extraction using MWPM decoding.

Highlights

• Deterministic surface-code simulations for distances d = 3, 5, 7
• Stable logical error rate (LER) estimation with confidence intervals
• Reproducible results via fixed RNG seeds
• Batched trial execution with live progress logging
• Non-monotonic LER detection warning (finite-trial variance)
• Automatic CSV export for threshold analysis

Verified Capabilities

• Reliable large-scale runs (≥2000 trials)
• Consistent cross-run reproducibility
• No hang at higher distances (d=7 stress-tested)

This release establishes a stable foundation for upcoming hybrid CV + discrete QEC development in v1.0.

v0.8: Full syndrome graph MWPM

• Build complete defect + boundary graph
• Add all-pairs defect edges
• Add defect-to-boundary edges
• Route corrections along lattice shortest paths
• Ensure syndrome reproduction correctness
• Integrate with shared weight interface (uniform|neural|llr)
• Add smoke tests for boundary and pair cases
• Preserve threshold runner and metadata logging

This release upgrades MWPM to a full syndrome graph implementation
suitable for research-grade surface code threshold studies."

v0.7.1 – LLR Physical Weight Formalism

Highlights

• Physically grounded LLR weight formalism
• Shared weight interface (uniform | neural | llr)
• Separate noise channel support (data / meas / idle)
• Weighted Union-Find support
• Metadata-rich threshold CSV logging
• OpenMP-parallel threshold runner

Scientific Impact

This release introduces a physically motivated log-likelihood ratio
weight model for surface-code decoding, enabling realistic modeling
of heterogeneous noise channels and improved reproducibility.

Compatibility

Backward compatible with v0.7.0 CLI interface.

v0.7 – Thread-Safe Union-Find Parity + Neural-Guided Weight Interface

This release introduces a major architectural upgrade to the Surface Code decoding framework.

New Features

• Thread-safe Union-Find decoder
• Eliminated shared mutable state
• Stable under OpenMP multi-threading
• Reentrant decode() implementation
• WeightField abstraction layer
• Pluggable edge-weight interface
• Uniform (baseline) weight mode
• Neural-guided weight mode (deterministic mock implementation)
• CLI support
• --weight_mode=uniform
• --weight_mode=neural

Architectural Impact

This release establishes a shared weight-field interface that will support:

• Neural-guided MWPM (v0.8)
• Log-likelihood reweighting
• Hybrid CV + discrete QEC roadmap

Stability

• Verified under --threads=8
• No decoder_fail_rate
• Regression-tested against v0.6
• Stable up to 5000 trials

v0.6 — Surface MWPM Decoder + Parallel Threshold Engine

LiDMaS+ v0.6

This release introduces a stabilized Surface Code MWPM decoder together with a fully parallelized threshold estimation engine.

Core Additions

• OpenMP-enabled Monte Carlo threshold trials
• Thread-safe surface code simulation harness
• Wilson confidence intervals for logical error rate
• Automatic threshold estimation via curve crossing detection
• Preliminary finite-size scaling fit extraction (p_c and ν)
• Decoder failure detection and diagnostic dump system

Stability Improvements

• Fixed OpenMP race conditions in MWPM integration
• Thread-safe random seeding per trial
• Graceful decoder failure handling (no global abort)
• Optional failure diagnostics (surface_decoder_failure_dump.txt)

Threshold Harness Features

• Parallel execution via --threads
• Automatic crossing estimation via --estimate_threshold
• Scaling-fit extraction via --scaling_fit
• CSV export of all surface threshold data

Example Usage

./lidmas --surface_threshold \
  --decoder=mwpm \
  --threads=8 \
  --trials=5000 \
  --estimate_threshold \
  --scaling_fit

v0.2 – Validated LDPC Belief Propagation Engine

Research-grade C++ implementation of LDPC belief propagation (sum-product and normalized min-sum) with validated monotonic FER waterfall behavior.

Features:

• PEG-generated (n=1000, m=500) regular LDPC codes
• Sum-product and normalized min-sum decoding
• Monotonic FER waterfall validation
• Sanity checks at p=0 and p=0.001
• Diagnostics: parity satisfaction rate, max-iteration hit rate
• OpenMP parallel Monte Carlo sweeps
• Reproducible CMake build

Validated FER curve:
p=0.06 → FER=0.01
p=0.07 → FER=0.115
p=0.08 → FER=0.355
p=0.09 → FER=0.755
p=0.10 → FER=0.965

Build:

mkdir build
cd build
cmake ..
make
./lidmas --bp=sum-product

Reproducibility Metadata
Tested on:

- macOS (Apple Silicon, Clang + Homebrew OpenMP)
- Expected to build on Linux (GCC/Clang) with OpenMP
- CMake >= 3.16
- C++17 standard
- OpenMP enabled for parallel Monte Carlo