v0.1.0 — Research Scaffold

v0.1.0 — Research Scaffold

The first public infrastructure release of StarCraft AI.

This release establishes the reproducible research environment before real-data world-model training begins.

Included

• structured PyTorch world-model scaffold;
• deterministic CPU/CUDA smoke training;
• CI quality gates and nightly multi-seed research matrix;
• CodeQL and Dependabot;
• current competitive-map metadata;
• high-MMR replay acquisition and pinned replay manifests;
• provenance-traceable real replay fixture;
• research, architecture, benchmark, model-card, and data-card documentation.

Research status

M0 is complete. M1 focuses on converting real replay execution into a versioned ObservationV1 + ActionV1 -> ObservationV1 transition dataset.

There is no competitive gameplay checkpoint in this release.