HumanSignal's Adala auto-corrects prompts using labeled data, but docs have pitfalls

HumanSignal, the company behind Label Studio, has released Adala, an open-source labeling framework that allows agents to iteratively refine their own prompts based on ground-truth labeled data. A developer reviewing the project found several discrepancies between the official documentation and the actual codebase. The pip package has not been updated since November 2023, the quickstart example uses label types that no longer match the current code, and the stated Python 3.8 support conflicts with the project's pyproject configuration. The author cross-referenced the GitHub repository and source code directly to identify and flag these inconsistencies. While the core concept of self-correcting prompt optimization is considered promising, users are advised to verify documentation claims against the source before following setup guides.