research-paper-submission

license	Apache-2.0
name	research-paper-submission
description	Take a formal result to publication when it sits between fields — choosing a venue and positioning against the right literature, finding prior art that uses none of your vocabulary, making imported machinery graspable, and structuring the paper the way the target community expects. Use when drafting or revising a paper for EC/AAMAS/CAV/S&P/OSDI/POPL-class venues, writing a related-work or contributions section, checking whether a term or result is already taken, or preparing a submission-readiness pass. NOT for doing the underlying proof, derivation or experiment; NOT for house voice on internal write-ups (use harbor-exposition); NOT for grant proposals, theses, or blog posts.
allowed-tools	Read,Write,Edit,Grep,Glob,Bash(python3:*),WebSearch,WebFetch
metadata	{"category":"Writing & Communication","pairs-with":"harbor-exposition, port-daddy-expository-writer, falsification-first","tags":"academic-writing, paper-submission, venue-selection, related-work, prior-art, exposition, figures, peer-review","version":"1.0.0"}

Research Paper Submission

Getting a between-fields result published. The failure mode this skill exists for is not bad writing — it is a correct, well-written paper that no community recognises as theirs, positioned against literature its author never knew to search.

When to Use

✅ Use for: choosing a venue; writing a contributions or related-work section; checking whether a term you are about to coin is taken; finding prior art in a field whose vocabulary you do not speak; making imported machinery readable for non-specialists; a pre-submission correctness pass.

❌ NOT for: doing the proof or experiment (do that first — this skill presents it); internal house-voice write-ups (harbor-exposition); grant proposals, theses, or blog posts; general copy-editing.

The premise that governs everything

Work at an intersection is usually transplantation: an established theorem from a community that has never heard of your application, re-derived against your scenario. That is a legitimate contribution, and it determines how you will be reviewed. It fails in three ways, and knowing which one you face decides every subsequent choice.

flowchart TD
  A[A result at an intersection] --> B{Do the two literatures<br/>cite each other?}
  B -->|yes| C[Inside an existing conversation.<br/>Risk: missing the closest MODERN instance,<br/>not the classical source.]
  B -->|no| D{Does the bridge import machinery<br/>that was unavailable before?}
  D -->|yes| E[A real bridge. Contribution is the identification.<br/>Risk: no PC knows both halves.]
  D -->|no| F[A relabel. Say so, or find the real delta.]
  C --> G{Do the paper's halves share<br/>any candidate venue?}
  E --> G
  G -->|no| H[Two papers. Split them.<br/>A shared slogan is not a shared contribution.]
  G -->|yes| I[Write the vocabulary section.<br/>Assume reviewers expert in one half, lost in the other.]

Order of work

Do these in order. Steps 1–2 change what the paper is; doing them after drafting means rewriting.

1. Positioning — fill in templates/positioning-worksheet.md before writing the contributions paragraph. It is built from real failures; every question corresponds to one.
2. Prior art — run the protocol in references/finding-prior-art.md. Snowball two rounds from a seed; cross vocabulary boundaries deliberately via controlled vocabularies; run the naming check on anything you plan to coin.
3. Venue — references/venue-map.md. Page budget differs by more than 2× across candidates, so a late choice means a rewrite.
4. Structure — references/exemplar-structures.md. Match the target community's dialect for contributions, related work, and where notation goes.
5. Exposition — references/exposition-craft.md. Definitions local and immediately before use; theorems readable cold; analogies that carry a candidate inference.
6. Figures — references/figures-and-examples.md.
7. Mechanical pass — scripts/submission_lint.py, then answer every claim-to-confirm it raises.

python3 scripts/submission_lint.py paper.tex --figures-dir ../figures
python3 scripts/submission_lint.py paper.tex --bib refs.bib --quiet-info   # errors only
python3 scripts/test_submission_lint.py                                     # 20 assertions

Anti-Patterns

Searching only in your own vocabulary

Novice: "I searched thoroughly and found no prior work." Expert: You searched thoroughly in your own words. A paper that solves your problem under another name is invisible to every keyword you would think of. The fix is structural: state your result with no term of art from the field you import from, then search that; use each adjacent field's controlled vocabulary (ACM CCS, MSC, JEL, arXiv categories); snowball forward from any one anchor you do find. Detection: your related-work section cites only venues you already read. A result at an intersection of n fields with citations from one field is not finished. Real cost: a characterization claimed as a contribution had been published 13 years earlier, with the same alphabet split and the same degenerate case, and was found only because an unrelated third paper mentioned it in passing.

Coining a term that is already taken

Novice: "I'll call this regimentation — it's descriptive." Expert: quote-search the exact phrase across adjacent fields before it enters the draft. If taken and it means the same thing, adopt and cite — that converts a novelty risk into a free citation of the founding work. If taken and different, rename. Coin only when neither applies. Detection: any term introduced without a citation that a reader in an adjacent field might already know. Real cost: one paper used a normative-multi-agent-systems term of art as if coining it; another named a theorem after a result in information economics that concludes the opposite.

Front-loaded preliminaries

Novice: "Section 2: Preliminaries. All notation and definitions." Expert: every practitioner source that can be verified disfavours it — Tsitsiklis marks it "optional; avoid it if you can", Krantz says it discourages readers, Dreyer's "just in time" is a structural argument against batching. What they do require is a tight local run of definitions immediately before the result that needs them. The difference is proximity, not whether definitions appear — the referees' most common complaint is still missing definitions. Detection: a definition in §2 whose first use is in §5.

Formalism without its intuitive reading

Novice: the theorem is stated precisely, so the paper is done. Expert: Tao's target is the post-rigorous stage — formalism present and checkable, always alongside the intuition. A paper showing only rigorous formalism asks every reader to redo the labour the author already did. Detection: a boxed theorem with no sentence before it saying what it means.

One example, mandatory for everyone

Novice: the worked example belongs in the main argument. Expert: two cognitive-load results pull opposite ways. The worked-example effect (Sweller & Cooper 1985) says a newcomer measurably needs it. The expertise-reversal effect (Kalyuga et al. 2003) says the same example taxes the specialist, who must reconcile it against a schema they already hold. So the example must exist and be visibly skippable — a labelled box or aside — so each reader takes only the load they need. Detection: a specialist cannot reach the theorem without reading the toy case.

Decorative analogy

Novice: "It's like a bouncer at a door" — vivid, so it helps. Expert: Gentner's structure-mapping says analogy transfers relations, not attributes, and earns its place by licensing a candidate inference — something you did not already know, projected from base to target, that then checks out. An analogy sharing surface features is a mere-appearance match and is "sharply limited in predictive utility." Detection: the analogy cannot survive one "so does that mean…?" question.

Trusting a verification sweep that cannot fail

Novice: "4,000 randomized instances, zero violations." Expert: check what the assertion actually evaluates. A sweep whose test rearranges the inequality it is testing, or whose loop never consults the mutant flag, is a tautology with a seed. It will report zero violations on a false theorem. Detection: can you algebraically derive the assertion from the theorem statement printed above it? Then it is not evidence. Real cost: a "76,000 schedules, 0 dominate" claim was entailed by an inequality chain three lines above it — and the sweep was therefore structurally incapable of catching the infeasible optimum it was meant to validate.

Superlatives the proof does not deliver

Novice: "unbounded depth", "provably does not", "enforces exactly". Expert: each is a promise with a specific proof obligation. "Unbounded" must not describe a logarithmic bound. "Exactly" is an iff and needs both directions, checked at the degenerate cases. A negative existence claim needs a proof, not a failure to exhibit. Detection: submission_lint.py flags all of these. It independently rediscovered four false "unbounded depth" claims and one false "provably does not" in a corpus that had already been manually audited.

Quality Gates

□ positioning-worksheet.md filled in BEFORE the contributions paragraph
□ Result stated once with no term of art from the imported field
□ Every coined term run through the naming check; decision recorded
□ Prior art searched in ≥1 adjacent field's own controlled vocabulary
□ Nearest prior work's theorem quoted verbatim WITH its hypotheses
□ Closest modern instance cited, not only the classical source
□ Venue chosen before drafting; page budget matches
□ Contributions/related-work format matches the target community's dialect
□ Definitions local, immediately before first use — no orphan preliminaries
□ Every theorem has an intuitive reading adjacent to it
□ Worked example present and visibly skippable
□ Every analogy licenses a checked candidate inference
□ Honest boundary in the paper's own voice, named as the venue names it
□ Figures survive greyscale; colour is never the sole channel
□ Every verification claim could actually have failed
□ submission_lint.py: zero errors, every claim-to-confirm answered

References

• references/venue-map.md — Read when choosing a venue or working out which community a result belongs to. Format/blindness table, the three transplantation failure shapes, live adjacent conversations.
• references/finding-prior-art.md — Read when writing related work, checking whether a term is taken, or claiming novelty. Snowballing, controlled vocabularies, per-tool blind spots, verification discipline.
• references/exposition-craft.md — Read when placing definitions, stating a theorem, or writing an analogy. Verbatim guidance from Halmos, Knuth, Krantz, Tao, Mermin, Peyton Jones, Dreyer, Tsitsiklis; the two cognitive-load effects; Gentner's systematicity test.
• references/exemplar-structures.md — Read when structuring the paper or writing the first sentence. Nine opening moves with verbatim examples; per-community conventions; two structures worth copying wholesale.
• references/figures-and-examples.md — Read when designing a figure or caption. What is experimental versus expert opinion versus craft lore; the Okabe–Ito palette; theory-figure conventions.
• templates/positioning-worksheet.md — Fill in before the contributions paragraph. Ten sections, each from a real failure.
• scripts/submission_lint.py — Run before every submission. Dangling refs and citations, duplicate labels, uncaptioned floats, missing limitations, unhedged superlatives.
• scripts/test_submission_lint.py — Run after editing the linter. 20 assertions including the \\{ brace regression.

Skill Bundle Index

Every file in this skill, and when to open it. Auto-generated; run scripts/index_references.py --fix.

root

• CHANGELOG.md — Research Paper Submission — Changelog — Built from five parallel primary-source research passes: venue CFPs fetched and read, fifteen recent strong papers read in full, practitione

references/

• references/exemplar-structures.md — How strong papers in this space are actually built — Read when structuring a paper, writing the first sentence, formatting a contributions list, or deciding where the running example goes.
• references/exposition-craft.md — Explaining formal ideas to readers who are not specialists in them — Read when drafting a section that imports machinery from another field, deciding where definitions go, stating a theorem, or writing an
• references/figures-and-examples.md — Figures, captions, and worked examples — Read when designing a figure for a theory paper, writing a caption, or deciding how to use a running example.
• references/finding-prior-art.md — Finding prior art across a vocabulary boundary — Read when writing a related-work section, checking whether a term is already taken, or trying to establish that something is novel.
• references/venue-map.md — Venue map and literature positioning — Read when choosing where to submit, or working out which community a result actually belongs to.

scripts/

• scripts/submission_lint.py — submission_lint.py - mechanical pre-submission checks for a LaTeX paper.
• scripts/test_submission_lint.py — Tests for submission_lint. Run: python3 test_submission_lint.py

templates/

• templates/positioning-worksheet.md — Positioning worksheet — Fill this in before writing the contribution paragraph, not after.