beibaidraco/maieusis

Maieusis is a Python tool built around one idea: before scientists analyze data, they first have to figure out what question is even worth asking. Most AI research tools jump straight to running an analysis once a question is given to them. Maieusis works one step earlier, helping researchers form and test good questions before any analysis happens. Given a dataset, some source papers, and an optional research direction, Maieusis studies how those papers arrived at their scientific questions, then turns those patterns into new candidate questions for the current dataset. Each candidate question then goes through a check against the real dataset, where the tool asks whether the needed information is actually present, whether a fair comparison can be made, and whether the question can be answered without changing what it originally meant. The output is something the authors call a Scientific Question Dossier: a readable record showing either a workable plan for a question, or an explanation for why the question was rejected, deferred, or flagged with a warning. Every step leaves a paper trail so a researcher can see how each conclusion was reached. The project has three working parts. A coding assistant, either Codex CLI or Claude Code, handles a sandboxed workspace with limited, read-only access to dataset files. Separate AI models handle reading papers and generating candidate questions. A Python core enforces rules, checks outputs, tracks sources, and limits how many calls and retries can happen. The README is explicit that this is a research preview, version 0.1.0, and that it does not judge whether a question is novel, important, or true. It does not run the actual scientific analysis or claim any findings are correct. It requires Python 3.11, 3.12, or 3.13 and separate API keys for different AI providers, since it deliberately avoids using the same provider for question generation and independent review. It is released under the Apache 2.0 license.