Chapter 1 — Reps

Eleven reps to make Week 1 muscle, not memory. They move from choosing the domain you will keep all semester, to reading papers like a scientist, to building the integrity habits — verified citations, a real portfolio, the paper started — that the rest of the course stands on. Every rep moves your project forward. The keyboard is the gym. Do not read these — run them.

Ground rules

• Pick your domain in Rep 1 and commit. Almost every rep this week feeds the project you keep for sixteen weeks. Switching domains later costs you a literature review.
• Verify before you cite. No reference enters your portfolio until you have confirmed it in dblp / Semantic Scholar / the publisher. This is non-negotiable from Week 1.
• Predict before you measure. Before you run a tool or score a paper, write down what you expect. The gap between prediction and result is the learning.
• AI policy (Phase 1): You may use AI to find and summarize, but you do the judgment yourself, and you verify everything the machine touches. Every rep that uses AI ends with an honest one-line AI usage note: what you asked, what it got wrong, what you verified. The human owns the verdict.
• Keep a reps.txt in your portfolio repo. One short written reflection per rep. This is graded thinking, not busywork — and it is your first commit history.

Reps 1–3: Choose the ground you will stand on

Rep 1 — Choose your domain and write the question

From the menu in §1.6 (AI · security · networking · SE · databases · HCI · graphics · AR/VR · robotics · cloud · HPC · architecture · data science), choose one domain. Then write a single sentence in the form: “Does [a modern AI method] improve [a specific outcome] in [my domain], measured by [a metric]?”

Domain:  __________
Question: Does __________ improve __________ in __________,
          measured by __________ ?

Reflection: Why this domain? Name one thing you already know about it and one thing you do not. Could your sentence be false? If not, sharpen it until it could.

Rep 2 — Sort claims: knowledge vs opinion

Write five sentences about your domain — mix on purpose. Then label each testable claim (could be falsified by an experiment) or opinion (cannot). For every opinion, rewrite it into a testable claim.

Reflection: Which was harder to write — the testable ones or the opinions? Most people find opinions flow easily and testable claims take work. Why is that, and what does it tell you about the discipline of research?

Rep 3 — Place your work on the research-type map

Using the table in §1.5, decide which type your project will most likely be: empirical, systems, theory, HCI/user-study, or measurement. Write the bar reviewers will hold that type to (e.g. empirical → baselines + multiple seeds + significance).

Reflection: Could your question be answered as more than one type? If so, which type is cheapest for you to do rigorously, given your time and compute this semester? Pick that one and say why.

Reps 4–7: Read like a scientist

Rep 4 — Find three recent papers (and verify them)

Find three recent (ideally last ~3 years) conference papers in your domain — at least one anchored to a real applied-AI result. Before anything else, run each citation through the verifier:

python3 code/verify_citation.py "Exact Paper Title Here" --year 2024

Reflection: Did all three resolve cleanly? Record the verified citation (authors, venue, year, DOI/arXiv id) for each in your reps.txt. A paper you cannot verify is a paper you cannot cite.

Rep 5 — Pass 1 on all three (five minutes each)

For each paper, do a strict five-minute Pass 1 (title, abstract, intro, headings, conclusion, references skim) and answer the five Cs: Category, Context, Correctness, Contributions, Clarity. Use code/paper-anatomy.txt as your map.

Paper 1 — Category: ___  Context: ___  Correctness: ___  Contributions: ___  Clarity: ___

Reflection: Which of the three earns a Pass 2 and which stops at Pass 1, and why? Being able to decline to read deeply is as much a skill as reading deeply.

Rep 6 — Full structural analysis of one paper

Take the paper that most resembles what you want to do. Fill out code/structural-analysis-template.txt completely — section map, the four pillars rated 1–5 with a cited line per rating, the applied-AI thread, and the gap mined from its Future Work.

Reflection: Quote the single sentence in this paper that points at an open problem. Could it become a practicum-sized question for you? If yes, draft that question; if no, say what makes it too big or too small.

Rep 7 — Identify each section in an unlabeled paper (the widget)

Use the Paper Anatomy Explorer widget below the chapter. First click through the labeled sample paper to see each section’s job. Then take its unlabeled-paper challenge: identify each part by function with the headings stripped. Record your score.

Reflection: Which section did you mislabel, and what fooled you — a Related Work paragraph reading like Introduction? A Discussion limitation reading like a Result? What does that confusion teach you about reading by function rather than position?

Reps 8–9: Build the integrity habits

Rep 8 — Stress-test an LLM’s citations

Ask an LLM for five references on a narrow topic in your domain. Run every one through code/verify_citation.py, then confirm the survivors by hand in dblp or the publisher.

for t in "Title 1" "Title 2" "Title 3" "Title 4" "Title 5"; do
  echo "=== $t ==="; python3 code/verify_citation.py "$t"
done

Reflection: How many of the five were fabricated, how many had a wrong year/author, how many were clean? Compare your count to the Walters & Wilder figures in §1.9. End with your AI usage note: what you asked, what it got wrong, what you verified.

Rep 9 — Score the four pillars, with evidence

Take any one of your three papers and score it on novelty, rigor, reproducibility, significance (1–5 each). The rule: every score must point at a specific line, table, or figure. “Feels rigorous” earns zero credit.

Rigor: 4/5 — reports 5 seeds with 95% CIs in Table 3 (cite it).
Reproducibility: 2/5 — no code or seeds released; data described only in prose.

Reflection: Which pillar was weakest in this paper, and is that weakness a flaw or a fair limitation given its type? A measurement paper with no novel method is not “low novelty” — it is playing a different game.

Reps 10–11: Stand up the portfolio and the paper

Rep 10 — Initialize the research portfolio

Create your semester-long Git repository (see Appendix A for the no-install GitHub path). Minimum structure, committed:

research-portfolio/
  README.txt          # your domain + your one-sentence question
  lit/               # comparison matrix lands here in Week 4
  notebooks/
  data/DATASETS.txt
  paper/             # your template goes here (Rep 11)
  reps.txt            # this week's reflections

Reflection: Paste your first commit hash into reps.txt. Why does a commit history matter for reproducibility specifically — what does it let a future reader (or you, in Week 11) reconstruct?

Rep 11 — Start the paper in a real template

Copy code/paper.tex into a new Overleaf project (or compile locally) and get it to build. Choose acmart or IEEEtran to match a plausible target venue in your domain. Fill in only the title and stub the abstract with Beck’s four sentences (problem / why / contribution / implication).

Reflection: Verify the literal acmart/IEEEtran version string in Overleaf’s package list and record it. Why start the paper now, with nothing to report yet? Answer in one sentence — it should mention the word “map.”

Done? One Last Thing.

A miniature of Project 1, end to end. In your portfolio repo, produce:

1. domain.txt — your chosen domain and your one-sentence, falsifiable, AI-meets-domain question.
2. One completed structural-analysis-template.txt for the paper closest to your intended work, with every citation verified and every pillar rating backed by a cited line.
3. A three-paragraph reflection.txt — what kind of research you intend to do, why this domain, and the one gap (from a real Future Work section) you might one day occupy.
4. A built Overleaf paper from code/paper.tex, committed under paper/.

If you can do this for one paper tonight, you can do Project 1. That is the whole craft in miniature: choose the ground, read like a scientist, verify everything, and start writing the paper from day one.

Up next: Project 1 — P1: Read Like a Scientist.