light-paper-drafting

Paper Draft Writing

General Principles

Organize the whole paper around one question:how to make reviewers believe this paper is worth publishing. Every time you write a paragraph, ask yourself "will reviewers believe it when they read this, and what will they ask?"

Research before writing: every claim must have a verifiable source before you write it down; any sentence without source support must be marked [MATERIAL GAP], and never fill in from thin air (anti-leakage protocol).brief-driven: a mediocre writer + a good brief beats a good writer with no direction—before writing each section, first state clearly "what is to be argued/which evidence is used/which paper is the benchmark."

Operating Modes (choose a tier first, don't run the full workflow every time)

full(full draft) / outline-only(outline + argument map only) / abstract-only(abstract only) / section-redraft(rewrite a section) / self-review(self-check only). For each mode's trigger/input/steps/output/integrity-gate tier, see references/operational_modes.md。

Asset Navigation

Structure Templates：templates/ six types of skeletons—01_imrad.md(IMRaD empirical) / 02_review_survey.md(review/systematic review) / 03_theory.md(theoretical method) / 04_case_study.md(case) / 05_policy_brief.md(policy brief) / 06_conference.md(conference). Each includes section order + word limit per section + [figure slot/table slot] + required statement slots.
End-to-end Example：examples/worked_example.md—a real "before/after" comparison of brief→configuration→contribution list→outline→argument map→Introduction draft (before = hitting every hallucination red line, after = compliant and explicitly marked [MATERIAL GAP]/[RESULT GAP]）。
Self-check Checklist：references/self_review_checklist.md—8 dimensions of quality + 7 types of failure modes, checkable, with two columns for each item: "what to look at/what triggers rewriting."
Integrity Gate：references/integrity_gate.md—claim sampling quota (draft ≥30%, final draft 100%) + citation verification actions (curl DOI/Crossref/arXiv/OpenAlex, endpoints already tested and HTTP codes recorded) + Chinese-language literature verification protocol without DOI(three-field comparison method for CNKI/Wanfang/VIP bibliographic records; Chinese-language literature must not skip the integrity gate just because it has no DOI; when writing, this skill is the interceptor, and the verification executor is the Chinese verification fallback checklist of m10 citation, with both sides using the same standard).
reporting Guideline Mapping：references/guideline_map.md—experimental/observational/systematic review/diagnostic/prediction/animal/case/qualitative → CONSORT/STROBE/PRISMA/STARD/TRIPOD/ARRIVE/CARE/SRQR.
Required Statements：references/mandatory_inclusions.md—Data Availability/Ethics/CRediT/COI/Funding + venue-specific AI usage statement templates.
Machine Checker：scripts/draft_lint.py—checks for residual GAPs, missing statements, SOTA sentences without significance, and extracts citations pending verification.python scripts/draft_lint.py <draft.md> [--final]; self-test python scripts/draft_lint.py --selftest(already run successfully). Human judgment takes priority, machine checking is the fallback.

Before Writing

Confirm the innovation points (m03/m04), results and highlights (m06), and target venue style (db01/db02). For the writing order, refer to a mature pipeline (ARS):

Paper Configuration: determine the structure type (IMRaD / review / theoretical analysis / case / policy brief / conference paper) and target venue.
Contribution list (about 3 items): As the pillar of the whole paper, every section must serve it.
Outline + Argument Map: First draw out the "claim → evidence → rebuttal" chain before you start writing.
Extract from db02 thestructural conventionsof this type of paper (section order, word/page limits, figure and table norms), without copying the content. For the venue template, be sure to use theversion from the target year(NeurIPS/ACL/IEEE templates change year by year; note the double-blind anonymity rules). When reading external PDF/DOCX/PPTX materials, you can use MarkItDown to convert them to Markdown before feeding them to the model (markitdown paper.pdf -o paper.md）。

Key points by module

Title: accurate + informative + reflects the innovation, avoid exaggeration.
Abstract: background → problem → method → key results (with numbers) → significance, 150–250 words.
Introduction: problem → gap → what this paper does → list of contributions → preview of results. The gap must be supported by the literature (each gap backed by 3–5 citations).
Related Work: organize by taxonomy, and at the end of each category point out "how it differs from this paper"; do not write a running account. This is the deciding factor in review, so be willing to invest in synthesizing the literature review.
Method: from intuition to formalization, with clearly defined equations, accompanied by a framework diagram (m09), reproducible.
Experiments: first explain the setup (data/baseline/metrics/implementation details), then the main results, ablations, sensitivity, generalization, robustness, corresponding to the m05 design. Use "data-driven placeholders" in the results paragraphs: explicitly write out the key numbers + significance (e.g., p<0.001) + comparison with literature baselines.
Results Analysis: explain why, cite the insights from m06, and acknowledge unfavorable results.
Discussion: significance, scope of applicability, and dialogue with existing work.
Conclusion: echo the contributions, do not introduce new content.
Limitations + Future Work: be honest, and show reviewer-friendliness.

When the paper "falls a bit short": tell the story honestly

Not every paper has top-conference-level innovation. When m04 rates it as merely "incremental", the core overlaps with prior work, or the main method fails,telling a good story ≠ exaggerating or fabricating—it means presenting the real contribution in the most persuasive framing while strictly upholding the integrity gate. This is a writing skill, not a fig leaf for academic misconduct. Establish two iron rules first:(1) Never misreport ② incremental as ① innovative(m04 has already checked for core overlap; misreporting will be caught);(2) Never hide failures/negative results. On top of that:

Reposition the contribution, rather than exaggerate it. If the core has been done by prior work → do not pretend to be the first; instead play the "breadth/rigor/unified comparison/reproducibility" card: explicitly acknowledge the prior work (set up a dedicated "Closest prior work" paragraph, explaining the delta sentence by sentence), and shift the selling point to "the first systematic comparison of X/Y/Z under a unified protocol" or "the first solid evidence with multiple seeds + paired statistics + effect sizes". What reviewers hate is "pretending not to see prior work", not "incremental but honest".
Turn negative results/failures into selling points. The main method didn't work, but "we tried the seemingly obvious shortcut, proved it doesn't work, and explained why" has scientific value in itself—it prevents later researchers from repeating the same mistake, and it can even reinforce the main conclusion in reverse (e.g., "there is no data-free shortcut, so data-driven calibration is necessary"). Give negative results their own section, explain the mechanism clearly, and provide a falsifiable explanation.
Narrow the framing to a scope that holds up. If you can't reach "we solved X," honestly write "we characterized the boundary/cost of X under condition Y." A small but true claim is harder to reject than a big but hollow claim. Shrink the title, abstract, and contribution list in sync to a scale the evidence can support.
Organize with a story arc, but attach evidence to every turn. A good story = tension (a neglected cost/contradiction) → escalation (how common and severe it is, with numbers) → turn (can it be fixed, and at what cost) → landing (actionable recommendations). Every link is backed by real experimental numbers, not adjectives.
Honesty buys trust, trust buys acceptance. Proactively write full limitations, acknowledge unfavorable results, and name prior work that is stronger than yours—reviewer friendliness directly affects the score. Once a hidden weakness is exposed, the credibility of the whole paper drops to zero.
Choosing the right venue is also storytelling. Solid incremental work submitted to a workshop / applied track / short paper is more likely to be accepted than forcing it into a top conference main track; when handing off to m13 venue-matching, honestly report "this is incremental," let it match the realistic tier, and don't submit to a venue doomed to reject you just for face. After finishing each module, immediately self-check with the m08 (polish) + m14 (mock review) perspectives. For the complete checklist, see references/self_review_checklist.md(8 dimensions + 7 patterns, each item includes "what triggers a rewrite").

Mock review 8-dimension self-evaluation checklist(ScholarEval dimensions, point out what can be improved item by item; relative self-check is more reliable than absolute scoring): Novelty / significance and impact / methodological rigor / clarity / reproducibility / support from experimental evidence / positioning relative to existing work / honesty about ethics and limitations.

7-category AI failure mode red-line self-check(derived from The AI Scientist's failure mode analysis; clear each item while drafting): M1 treating an implementation bug as passing; M2 hallucinated citations (fabricated references); M3 hallucinated experimental results; M4 taking shortcuts; M5 packaging a bug as a "new insight"; M6 methodological fraud; M7 early frame-lock. Citation checking: citation key/DOI/arXiv ID are only provenance fields and do not mean the paper actually supports your argument—click through anything doubtful. For the checking quota and curl actions, see references/integrity_gate.md。 Use machine-readable citekeys uniformly for in-text citation placeholders: write it as \cite{authorYearWord}——first author surname + year + first content word of the title, all lowercase(e.g., \cite{zhang2024deep}); do not use [Zhang 2019] and similar free-text placeholders (when typesetting at m10 they won't match the .bib keys and will report undefined citation). This formula has the same source as the citekey pinned when m10 citation generates the .bib. Deliver only after revising. During revision, track the score dimension by dimension, and watch for regressions where "fixing A breaks B."

Output

Modular Markdown/LaTeX draft + contribution list + markers for content to be added ([MATERIAL GAP]/[RESULT GAP]/TODO, e.g., missing figures or experiments). Before submission, go through references/mandatory_inclusions.md 's declaration checklist and references/integrity_gate.md 's final-manuscript gate (100% claim/citation verification). Align terminology with the db09 glossary (a07). Put outputs in a timestamped directory for version management.

Handoff

Figures and tables go to m09/m11, citations go to m10 (in the main text, \cite{} placeholders use authorYearWord formula, from the same source as the citekey of m10 pin), typesetting goes to m12, polishing goes to m08, simulated review goes to m14. Before delivery, pass the a08 (light-self-review) self-check gate. The version goes into db09.

Tool research notes (real endpoints/methods/limitations) are in references.md。