Chapter 16 · Reps

The Research Symposium — Reps

← Back to Chapter 16

Chapter 16 — Reps

This week the reps are not warm-ups. Each one moves your own paper across the finish line — talk, finished manuscript, and a concrete publication roadmap to a real venue. Do them in order; they build the final.

Ground rules.

  • Work on your paper, the one you have carried since Week 1 — not a toy.
  • Open-AI for the work; log substantive AI use in agent-log.txt as you go (Appendix C). The model never bears the accountability — you do.
  • Every number you put on a slide must be in the paper. Every citation you keep must be verified against a real index.
  • Hedge time-sensitive facts “as of 2026” and verify any venue policy live before you rely on it.
  • Keep everything in your portfolio repo. Tag a release at the end of the week.

Rep 1 — Write the one remembered sentence

Before any slide, write the single sentence you want the audience to repeat after your talk. Format: “In [domain], [your AI method] [does what] compared to [baseline], measured by [metric].” Keep it falsifiable and specific.

One sentence: ______________________________________________________________

Reflection (2–3 sentences): If a listener could only remember this one sentence, would it be true to your strongest result and only that result? Cut any clause the data doesn’t fully support.

Rep 2 — Build the 12-slide talk

Fill out code/talk-outline.txt one beat per slide. Build the deck (slides, Google Slides, or LaTeX Beamer — your choice).

cp code/talk-outline.txt talk/outline.txt
# then build slides/ from the outline, one slide per beat

Reflection: Which beat was hardest to fit in its time budget — the approach (slide 5) or the result (slide 7)? Name what you cut, and confirm the cut detail still lives in the paper.

Rep 3 — Rehearse on a clock, twice

Present the full talk out loud, timed. Then do it again.

Run 1 time: ____  Run 2 time: ____   Target: <= 12:00 speaking
Slides cut after run 1: ____________________________________________

Reflection: What did the second rehearsal fix that the first one couldn’t? (It is almost always pacing on the result slide and a too-wordy setup.)

Rep 4 — Prepare three backup slides for the hardest questions

List the three hardest questions a reviewer could ask (baseline fairness? seeds? contamination? a confound?). Build one backup slide per question.

Reflection: For each question, write the one-sentence answer you’ll give if you don’t have a slide ready — including, where it’s the honest answer, “I didn’t test that; it’s future work.”

Rep 5 — Audit every citation

Export your Zotero library to BibTeX, then build the verification ledger and fill it by hand.

python code/citation_audit.py references.bib
# open citation_audit.csv; verify EVERY row in dblp / Semantic Scholar / publisher

Reflection: How many rows could you not immediately confirm? Any reference you cannot verify gets fixed or removed before submission — a hallucinated citation can get a paper rejected without review.

Rep 6 — Format the paper in the venue template

Move the draft into acmart (\documentclass[sigconf]{acmart}) or IEEEtran (\documentclass[conference]{IEEEtran}) on Overleaf. Make it compile clean and fit the page limit in the venue’s count format.

Reflection: Did formatting expose any figure that’s unreadable at column width or any section that’s over budget? List what you tightened (not what you added).

Rep 7 — Read a real CFP and fill the venue-fit sheet

Choose a real candidate venue (a workshop or Student Research Workshop is the honest practicum target). Read its actual Call for Papers. Fill code/venue-fit-sheet.txt — quote the scope line, the page limit, the deadline, the anonymity rule, and where AI disclosure goes.

Reflection: Quote the one CFP sentence that best matches your RQ. If you can’t find one, that’s a fit problem — name a better-matched venue.

Rep 8 — Check the venue’s prestige and AI policy

For your chosen venue, find its CORE rank (core.edu.au) and Google Scholar Metrics h5-index, and read its current AI-disclosure policy.

Reflection (3–4 sentences): Is this venue reachable for a WIP paper this cycle? Where exactly does its policy say AI disclosure goes — and does that match where you put yours?

Rep 9 — Write the honest gap assessment

Open code/submission-roadmap.txt and complete Phase 0. Be specific about the gap between your draft and a submittable paper: result strength (multi-seed? effect size? CI?), baselines, related-work currency, reproducibility, page fit, scope honesty.

Reflection: Name the single biggest gap and the smallest concrete step that would close it. This is the judgment the final is graded on — an AI can’t write it for you.

Rep 10 — Build the reproducibility artifact

Pin the environment, confirm seeds are set and data is documented, tag a Git release, and (if you have an account) deposit to Zenodo for a DOI.

pip freeze > requirements.txt        # or: conda env export > environment.yml
git tag -a v1.0-symposium -m "Symposium submission artifact"
git push --tags
# optional: connect the repo to Zenodo and publish the tagged release for a DOI

Reflection: Could a stranger, with only your repo and README, rerun the headline result? Name the one thing still missing if the answer is no.

Rep 11 — Write the AI-disclosure paragraph

Draft the exact disclosure paragraph for your paper, in the location your target venue requires (Acknowledgements for ACM/IEEE/ACL; experimental setup for NeurIPS if non-standard; both text and form for ICLR). Name the system and the level of use.

Reflection: Apply the assistive-vs-generative heuristic to your own work: which AI uses needed disclosure, and which (grammar, editing) did not? Confirm there is no hidden prompt text anywhere in the file.

Done? One Last Thing.

Capstone rep — the dry-run symposium.

Put it all together as if today were the symposium. In one sitting:

  1. Give the full 12-minute talk out loud, on a clock, to a friend, a mirror, or a recording.
  2. Have them (or yourself) ask your three hardest questions; answer without bluffing.
  3. Open the formatted paper and confirm: every slide number appears in it, every citation is verified, the disclosure paragraph is in the right place, no hidden text.
  4. Read your completed code/submission-roadmap.txt aloud and check that it names a real venue, a real deadline, and a real continuation plan.

Reflection (a paragraph): Standing at the end of sixteen weeks — is this work you can stand behind in front of a room? If yes, you’re ready for the final. If not, name the one thing to fix, and fix it. That honest answer is the whole point of the week.

Up next: Project 14 — the FINAL.