Chapter 14 · Reps

Writing the Research Paper — Reps

← Back to Chapter 14

Chapter 14 — Reps: Assemble the Draft Paper (Graded Lab, 10%)

This is a drill week, so the reps are the graded deliverable. By Friday you assemble everything you have built since Week 1 into one complete conference-style manuscript — the Draft Paper (10% of your final grade) — in a real publication template. It does not have to be polished. It has to be complete and honest: every section drafted, every citation verified, the Methods/Results/Discussion firewall intact, so a peer can review the whole arc in Chapter 15.

“Write the vision; make it plain on tablets, so he may run who reads it.” — Habakkuk 2:2 (ESV)

Submit: a compiled PDF + the Overleaf/LaTeX source + the updated references.bib, committed and tagged in your portfolio Git repo (git tag draft-v1). Allowed tools: anything — Overleaf, your editor, AI drafting assistants. AI policy: open for drafting and editing, but you own every word, figure, and citation; verify every reference by hand (no hallucinated sources); if substantive text is AI-generated, note where, ready to disclose per your target venue (Appendix C). “The model did it” is never a defense.

Ground rules

  • Use your real research, not a toy. This draft is the spine of your final paper (Chapter 16); every rep moves your own project forward.
  • Compile early, compile often. Get a one-page skeleton building before you fill it. A first-time LaTeX error on deadline night is self-inflicted.
  • Firewall discipline: Methods = what you did; Results = what you found (no spin); Discussion = what it means. Never smear them.
  • Honest record: every run you ran goes in. Dropping inconvenient runs is falsification, not editing.
  • Verify every citation against dblp or Semantic Scholar before it enters references.bib.
  • Each rep ends with a short written reflection — keep these in portfolio/week14-reflections.txt; your reviewer and your instructor both read them.

Rep 1 — Stand up the template and make it compile

Choose the template that matches the venue you are aiming at (you will lock it in Chapter 16): ACM acmart for ACM venues, IEEE IEEEtran for IEEE. On Overleaf, start from the official template (Templates → search “acmart” or “IEEEtran”). Drop in code/paper-skeleton.tex as your structure, set the submission class, and compile a one-page skeleton with all section headings and placeholder text.

% ACM submission, single-column, anonymous for double-blind review
\documentclass[manuscript,review,anonymous]{acmart}
% --- or IEEE ---
% \documentclass[conference]{IEEEtran}

Reflect: Which template, and why does it fit your target venue? Did it compile clean on the first try — and if not, what broke?

Rep 2 — Write the abstract with Beck’s five moves

Draft the abstract (150–250 words) using the five moves: context → gap → approach → result → significance. Paste it into the Paper Structure Builder widget on the chapter page; it will not mark the abstract complete until all five moves are present, and it warns if your result move has no number. Put a real number with its uncertainty in the result move.

[CONTEXT]      Modern <domain> increasingly relies on <AI method>...
[GAP]          ...but <the specific thing no one has measured/done>.
[APPROACH]     We <your contribution>, evaluating on <benchmark/data>.
[RESULT]       We find <number + 95% CI / effect size>.
[SIGNIFICANCE] This bounds <where it can/can't be trusted>.

Reflect: Which move was hardest? (For most people it is the third — the startling sentence that makes your paper different from every other.)

Rep 3 — Introduction via CARS, ending in a contributions list

Write the introduction with Swales’ three moves — establish the territory, establish the niche (the gap, backed by a cited absence), occupy the niche — and close with an explicit bulleted contributions list (C1, C2, C3…). If you cannot write three concrete, checkable contributions, your claim is not sharp enough; revisit your research question (Chapter 6) before continuing.

Reflect: Write your gap as one sentence beginning “However, prior work…”. Then write the one sentence beginning “In contrast, we…”. If those two sentences don’t lock together, your paper isn’t focused yet.

Rep 4 — Related Work from the comparison matrix

Convert your comparison matrix (Chapter 4; template in code/comparison-matrix.csv) into a positioned Related Work section. Group prior work by approach, not by paper; name what each camp does and where it falls short; then position yourself. Represent every competitor at its strongest — no strawmen (2 Cor. 13:1).

Reflect: Which column of your matrix is empty across all rows? That empty column is your gap — state it in one sentence here.

Rep 5 — Methods, written for a replicator

Draft the Methods section from your experimental-design and reproducibility work (Chapters 7 and 9). Include: data and splits, baselines and ablations, the metric and why it operationalizes your construct, number of seeds/runs, hardware, and software versions. If an LLM was an instrument, record the model name, version, date, temperature, and prompt (prompt can go to an appendix). Describe the final clean procedure — not your debugging journey.

Reflect: Hand your Methods to a labmate and ask: “Could you rerun my study from this alone?” Record what they said was missing.

Rep 6 — Results, behind the firewall

Draft the Results section: tables and figures carry the load, prose points to them. Report effect sizes and confidence intervals, not bare p-values (Chapter 12). If results come from multiple seeds, report the spread (Bouthillier et al. 2021). Every run goes in. Then audit it: read each sentence and ask, “Does this require interpretation?” If yes, it belongs in Discussion — move it.

Reflect: List any sentence you had to move from Results to Discussion. How many were there?

Rep 7 — Figures and captions that stand alone

Produce at least one figure and one table for the paper. Export figures as vector PDF/SVG (plt.savefig("fig2.pdf")), put table captions above and figure captions below, show uncertainty (error bars / CI bands, stated in the caption), and use honest, non-truncated axes. Reference each in the text, in order, before it appears.

Reflect: Cover the body text and read only your figure + caption. Does it stand alone? Rewrite the caption until it does.

Rep 8 — Discussion and Conclusion

Draft the Discussion: interpret your results, relate them to prior work (back to the matrix), state limitations and threats to validity (map to Cook & Campbell’s four categories — statistical conclusion, internal, construct, external), and do not overclaim relative to what your CIs and effect sizes actually support (ASA 2019). Close with a Conclusion + Future Work that states the takeaway and the open doors without hedging your own contribution into nothing.

Reflect: Write one sentence per validity category naming a real threat to your study. If any category is empty, you haven’t thought hard enough.

Rep 9 — The citation hygiene audit

Open references.bib (start from code/references.bib) and verify every single entry against dblp or Semantic Scholar — especially any reference you first met through an AI tool. Confirm authors, venue, and year; copy the BibTeX from the real index, not from a chatbot. Delete anything you cannot resolve.

# For each \cite key in your paper, confirm the source exists on dblp.
grep -oE '\\cite\{[^}]+\}' *.tex
# Then search each key's title at https://dblp.org and https://www.semanticscholar.org

Reflect: How many entries had a metadata error? Did any fail to resolve at all? (Recall: ~18–55% of LLM-produced citations are fabricated, ~87% of real ones carry metadata errors — Walters & Wilder 2023; Bhattacharyya 2023.)

Rep 10 — Pre-flight and reproducibility check

Run the pre-submission checklist in code/repro-checklist.txt: every section present, every figure referenced, every citation verified, the firewall clean, env/seeds/data documented, AI use noted for disclosure. Confirm the PDF compiles from a clean checkout of your repo.

Reflect: What did the checklist catch that you would have missed?

Done? One Last Thing.

This is the capstone rep — it rehearses next week’s peer review and the final paper.

Read your own complete draft, end to end, in one sitting, as a hostile reviewer. Then write a 200-word self-review answering three questions:

  1. What is the single weakest claim in the paper — the one a reviewer will attack first — and is it backed by the evidence you actually have?
  2. Is the Methods/Results/Discussion firewall clean anywhere it matters, or did interpretation leak into Results?
  3. Could a stranger rerun this from the Methods section and your repo alone? If not, name exactly what is missing.

Commit and tag the draft (git tag draft-v1). You will hand it, anonymized, to two classmates in Chapter 15 — Peer Review and Revision. Make the vision plain. Then let someone run with it.

Up next: Chapter 15 — Peer Review and Revision — two classmates review your draft under double-blind rules, and you review theirs.