Skip to content

MPSBoost 0.4.0 Release Audit

This document records the release gate for 0.4.0: the 0.x release after the large real-world acceptance pass. It does not rewrite the 0.3.0 audit; 0.3.0 remains the all-trees feature milestone before the large-scale S18 validation pass.

Scope

0.4.0 includes the 0.3.0 all-trees scope plus:

  • optional official-SHAP integration diagnostics with native tree export payloads;
  • explicit portable-backend diagnostics and optional extras for XGBoost, sklearn, and CUDA;
  • project-local opt-in real-world dataset caches for MNIST, Titanic, Adult Income, Covertype, and HIGGS;
  • the S18 real-world acceptance report with success, degradation, and boundary notes;
  • documentation that 1.0.0 remains blocked until final stable-release gates are complete.

Boundaries

Native CPU/MPS remains the MPSBoost implementation and correctness oracle. Linux CPU and Linux CUDA execution are exposed through explicit external backend policy and adapters; failures inside the external sklearn/XGBoost/CUDA runtime are not represented as native MPSBoost backend failures.

1.0.0 remains blocked until model-quality, end-to-end performance, peak-memory, wheel/model-size, permission, artifact-hash, and explicit user-confirmation gates are complete.

Validation Matrix

Required before publishing 0.4.0:

  • local non-GPU unit, integration, packaging, and real-world acceptance tests;
  • opt-in external real-world dataset acceptance from project-local caches;
  • GitHub hosted CPU/package tests for Python 3.10 and 3.13;
  • self-hosted real Metal GPU tests for Python 3.10 and 3.13;
  • twine check for the exact uploaded wheels;
  • fresh wheel install and import/training smoke verification.

Artifact Rules

The release wheel must not contain specs/, tests/, benchmarks/, .github/, build directories, cache directories, raw datasets, credentials, or runner files.