paper-structure

Paper structure

Organization, not sentences. This skill decides what goes where and how much room it gets. For how a sentence sounds, use humanize with the register the venue file names. For whether a number is trustworthy, use sensemaking. For whether a claim outruns its evidence, use skeptical-labmate.

Structure is not universal. A clinical physics manuscript puts Methods before Results and weaves prior work through the Introduction and Discussion; a machine learning conference paper gives Related Work its own section and wants a reproducibility statement; some high-profile journals put Methods last. So the skeleton comes from the venue, never from habit.

Start by loading the venue

Read venues/<venue>.md before anything else. If the venue has no file, write one from venues/_template.md first — it takes fifteen minutes and every later decision depends on it. If the venue is genuinely undecided, pick the most likely one, say so, and expect to restructure.

Check the verified date in the file. Limits and required sections change between cycles. If it is more than a year old, confirm the numbers against the venue's own page before trusting them for a submission.

Fix the spine before writing anything

Write one sentence that is the paper's claim. Not the topic, not the area: the thing a reader should believe afterwards that they did not believe before.

Every section then earns its place against that sentence. A section that does not advance it, support it, or bound it is cut or demoted to supplementary material, however much work it represents.

Two failure modes this catches early:

• Two spines. If the sentence needs an "and", you may have two papers. Not always — a method and its validation are one spine — but check, because a two-spine paper reads as unfocused at every venue and reviewers say so.
• A spine that is a description. "We built a pipeline for X" is a topic. "X is determined by Y, not by Z" is a claim. The second can be wrong, which is what makes it worth reading.