v0.1.0-alpha — public alpha for repo continuation evaluation

RepoScale is now public as an alpha benchmark for repo continuation intelligence.

This release packages a runnable evaluation pipeline for measuring how well models understand and continue existing software projects instead of isolated coding tasks.

What ships in this alpha

• 12 curated cases across Diagnose, Intent, and Plan
• CLI workflow: validate, batch, score, summary, compare
• Structural, heuristic, and LLM-judge scoring
• Judge stability support with repeated scoring
• Corpus manifest, case authoring guide, and baseline scripts
• GitHub Actions CI

Good use cases right now

• Evaluate models on repo understanding and continuation-style tasks
• Run small reproducible baselines across providers
• Contribute new cases and scoring improvements

Important alpha caveats

• Methodology is still evolving
• Judge neutrality is not yet benchmark-grade final
• Corpus size is useful but still small

Start here:

• README: https://github.com/mverab/Reposcale
• Corpus: https://github.com/mverab/Reposcale/blob/main/cases/CORPUS.md
• Case authoring guide: https://github.com/mverab/Reposcale/blob/main/docs/case-authoring.md