xueruisu/paper-prism/research-understanding

Research Understanding

Read ../_shared/paperprism_codex_native.md before acting.

Goal

Turn messy research material into a readable ResearchAsset without requiring fixed JSON or Python execution.

Inputs

• Idea, draft, LaTeX, table, experiment logs, related-work notes, review comments, or conversation.
• Optional existing ResearchAsset.md or paper draft.

Workflow

1. Extract task, method, insight, evidence, baselines, related works, theory, limitations, and target venue.
2. If the user provides a draft, extract assets before rewriting.
3. Update ResearchAsset.md or provide an equivalent concise working summary.
4. Identify the top missing inputs that block claims or writing.
5. Ask at most three high-value questions; otherwise continue with caveats.

Rules

• Do not invent results, baselines, citations, theory, or related-work facts.
• Do not ask the user to fill a schema.
• Preserve uncertainty and source provenance.
• Treat incomplete material as usable but risky, not as a reason to stop.

Expected Outputs

• Updated or proposed ResearchAsset.
• Top missing information.
• Handoff recommendation to framing, claim-evidence mapping, or writing.