Research Best Practices

Your AI builds it with you — just work. It drafts onto your proofline as you research; you curate the pending cards. If it isn't capturing, tell it to "use Proofline" / "capture this to my proofline."

Captures are drafts on your proofline — accept, edit, or reject them in the chat ("what's pending?" → "accept this") or on the web. Nothing commits without your say-so.

Push your own documents in — and cite them. Ask your AI to add a document you have — a PRD, a methods note, a paper — to your library, then cite it on your findings; a source in your own library grounds a claim to full Support (higher than an external link).

Cite with a real source, never a bare title. A DOI, a paper link, or one of your own library documents — a bare title doesn't count toward Support.

Abstaining is a result, not a failure. "Not enough evidence yet" is the right, honest answer when evidence is thin.

Ask "where were we?" to re-orient — your AI reads your actual proofline, not its memory. Any session, any device, any tool.

New prooflines start private to you. Tell your AI to share one with your team and everyone in your workspace can open and build on it — including anyone who joins later. If a teammate can't see a proofline, it's probably still private — just share it.

One line per direction; close each with its own conclusion. If you're exploring a few angles, have your AI open a line per angle first — that keeps the structure clean.

A conclusion closes its line; a conclusion drawn across lines is your overall answer. That's the shape of a finished proofline.

Notice a conflict? Resolve it. Tell your AI two findings conflict; once you've worked out the tension, ask for a conclusion that draws on both — the conflict resolves itself.

The why — questions, evidence, claims, conclusions. Your proofline is a reasoning record, not a status board: capture the why, keep status in your tracker. And use it for any structured reasoning — a question, a decision, a plan — not just literature review.