Running Research

With AI agents writing analysis scripts and data pipelines, research is code work — and it has the same coordination problems: what should the agent do next, what did we already decide, and who checked this result. Uclusion gives a solo researcher the same answers it gives a solo developer.

Hypotheses are options on a question

Create a question for anything genuinely open — “Which model family fits this data?”, “Is the effect real or an artifact?” — and use Add Option to enter each hypothesis.

As evidence arrives, vote on options with a certainty and a written reason, or comment inside an option with what you found. New hypotheses become new options at any time. When you finally decide, resolving the question freezes the full record: the winner, the rejected options, and every reason — readable months later by you or by an AI agent starting a fresh session.

Questions stay open as long as they need to. A question that waits on an experiment is normal, not overdue.

Let your AI do the work — with oversight

Connect your AI and mark work “AI able”. The agent finds its own work, asks questions with options instead of guessing, approves what it plans to do with a certainty score and written reason, and asks for review when results are ready. Nothing an agent produces has to enter your record unreviewed.

Everything else you already do

• Track experiment runs and analysis chores as jobs and tasks, with stages making the state of every strand visible at a glance.
• Keep lab-notebook style notes at the view or job level — decisions and context accumulate on the work itself as it happens, and the Show AI checkbox controls what your agent sees.
• Hand results to an advisor or collaborator with a generated Markdown report — no account needed to read it — or invite them as a free observer.
• Your notes stack stays. Zotero, Obsidian, and Overleaf are unaffected; Uclusion holds the questions, evidence, and decisions above them, and the CLI can export the whole workspace to Markdown at any time.