Start with a research question, then build a manuscript around evidence.
A good paper is a controlled argument. It explains a specific problem, shows why existing work is insufficient, presents a method, and proves the contribution with evidence.
gumu is built around the full paper object: prompts, chats, source edits, PDF regions, references, experiments, and venue targets.
Name the system, population, or phenomenon; the gap; and the measurable outcome. A question that cannot be tested will produce a vague paper.
Use the manuscript starter to turn a rough idea into candidate questions, hypotheses, and venue fits.
Write one sentence for what is new, one for why it matters, and one for how you validate it. Keep these sentences visible while drafting.
Ask gumu to pressure-test contribution claims against related work and likely reviewer objections.
Use the standard paper arc only if it serves the evidence: introduction, background, method, evaluation, discussion, limitations, related work, conclusion.
Generate a venue-specific outline and ask which sections need figures, tables, or formal definitions.
Every paragraph should advance a claim, define a concept, report evidence, or connect the paper to prior work. Remove paragraphs that only sound academic.
Run section passes that ask whether each paragraph has a clear role in the argument.
Writing exposes missing baselines, weak metrics, and unclear hypotheses. Update experiments and text iteratively instead of waiting for a perfect result table.
Keep experiment notes and manuscript edits in one thread so changes to results propagate into the paper.
Before submission, simulate a critical review: overclaims, missing baselines, ungrounded related work, unclear figures, weak limitations, and artifact gaps.
Run the paper review agent, reference verifier, figure pass, and camera-ready polish pass before exporting.
Start from a rough prompt, paper draft, PDF, review, call text, or supervisor note. gumu keeps the chat, source, references, and PDF regions together so every answer can become a concrete edit.
Write a one-page skeleton: question, contribution, closest prior work, method sketch, evaluation plan, and expected limitations.
Draft it early to clarify the story, but rewrite it after results and limitations are final.
It is ready when a critical reader can identify the problem, contribution, evidence, and limits without asking you to explain the paper orally.