Chapter 8 — Reps

This week the reps are not warmups around the project — they are the project, built piece by piece. Every rep moves your own research proposal forward. By the end, the talk, the preregistration, and the revised draft exist.

Ground rules

• Work on YOUR project. Every rep operates on your chosen domain and your real RQ — not a toy.
• Verify every citation. Anything an LLM hands you gets checked against dblp or Semantic Scholar before it enters a document. No exceptions (55%/18% — Walters & Wilder, 2023).
• AI is a whetstone, not a judge. Use it to sharpen; you decide what’s true and you own every claim. Disclose substantive use per your target venue.
• Commit as you go. Each artifact lands in your Git portfolio with a dated commit message. The portfolio is the record.
• Write reflections in plain prose. A rep isn’t done until you’ve written the 2–4 sentences it asks for.

Rep 1 — Assemble the seven-part skeleton

Pull your first-half artifacts into one proposal document with the seven section headers: problem & motivation, related work, research question, method, validity & threats, expected outcomes, timeline & feasibility.

proposal.docx
├── 1. Problem & Motivation      (from Ch2 + the Ch4 matrix)
├── 2. Related Work               (from Ch3/Ch4 lit review)
├── 3. Research Question          (from Ch6)
├── 4. Method                     (from Ch7 design doc)
├── 5. Validity & Threats         (Ch6 — Cook & Campbell ×4)
├── 6. Expected Outcomes          (NEW this week)
└── 7. Timeline & Feasibility     (NEW — Weeks 9–16)

Reflect: Which section is thinnest right now? That thin section is where a reviewer will push. Name it in one sentence.

Rep 2 — The falsifiable claim, in four lines

Write, verbatim, four lines: your research question (one sentence), H1, the null H0, and the exact result that would refute H1.

RQ:        ________________________________________________
H1:        ________________________________________________
H0:        ________________________________________________
Refuted if: _______________________________________________

Reflect: Read “Refuted if” out loud. If you can’t imagine that outcome happening, your hypothesis isn’t falsifiable yet — rewrite it until the refutation is plausible.

Rep 3 — Preregister the plan

Fill code/preregistration.txt completely and post it to OSF or AsPredicted.org (both free, as of 2026). Paste the time-stamped link into your proposal.

Reflect: Which choice did preregistering force you to make now that you’d been leaving vague — the metric, the seed count, the stopping rule, or the exclusion rule? Why was it tempting to leave it open?

Rep 4 — Name the baseline that hurts

In your method section, name the single baseline your method must beat — and choose the one whose defeat would mean something, not a strawman.

Baseline:                 ____________________
Why beating it matters:   ____________________
The ablation testing my key claim: ____________________

Reflect: Would you be comfortable if your method lost to this baseline and you had to report it? If not, you’ve picked a baseline you’re afraid of — which is exactly the right one.

Rep 5 — Build the 12-slide talk

Using code/proposal-slide-outline.txt, build the deck. One idea per slide, RQ on slide 1 and slide 6, the comparison matrix on slide 4. Then rehearse out loud, on a timer, three times.

Reflect: After the third rehearsal, which sentence did you hear yourself say that you don’t fully believe? Write it down — that’s where Reviewer 2 is going.

Rep 6 — The threats-and-mitigations slide

Build slide 10: your top two validity threats (mapped to Cook & Campbell’s four categories — statistical conclusion / internal / construct / external) and a concrete mitigation for each, plus your one-line AI-disclosure plan.

| Threat (category)        | Mitigation                |
|--------------------------|---------------------------|
| ____________ (________)  | _________________________ |
| ____________ (________)  | _________________________ |
AI disclosure: tool ____ used for ____ ; disclosed in ____

Reflect: Which threat is hardest to mitigate within the remaining weeks? Honesty here is worth more than a clever-sounding fix you can’t actually do.

Rep 7 — Score a peer with the rubric

Take a classmate’s draft proposal (or a sample). Score it on the six dimensions in code/proposal-peer-review-rubric.txt out of 35, then write the one constructive question that would improve it most — specific, cited to a slide/line, answerable.

Reflect: Which was harder: assigning the scores, or writing the one good question? The question is the skill that transfers to every venue you’ll ever review for.

Rep 8 — Run the LLM as Reviewer 2

Prompt an LLM: “You are a skeptical Reviewer 2. Here is my RQ, hypothesis, and method. Ask the five hardest questions, and for each name the validity threat behind it.” Capture the five. Then verify any citation the model produced against dblp or Semantic Scholar.

Reflect: How many of the five questions did your draft already answer? Of the citations the model touched, how many were real and correctly described? Report the count honestly — that number is your hallucination rate for this session.

Rep 9 — The revision memo

After your peer-review session (real or simulated with Rep 7/8), triage every comment into major / minor / clarification and write a one-page revision memo: for each comment, what you changed, or why you disagree (with evidence).

| # | Comment | Bucket | Action / Rebuttal |
|---|---------|--------|-------------------|

Reflect: Which comment generated the most heat in your chest? Re-read it. The hard review is usually the correct one — what is it actually pointing at?

Rep 10 — Midterm methods warm-up

From code/midterm-review-map.txt, cover the right two columns and, for each week 1–7, state the testable idea and name the anchor from memory. Check yourself against the map.

Reflect: Which two weeks did you blank on? Re-read those chapters’ “Why This Matters” openers before the Canvas quiz.

Done? One Last Thing.

The full dress rehearsal. Put it all together end to end, as if the room is watching:

1. Deliver the 10–15 minute talk on a timer (Rep 5), no notes on the slides themselves.
2. Have a peer (or the LLM-as-Reviewer-2 from Rep 8) fire the five hardest questions. Answer each — and when you don’t know, say “I don’t know; here’s how I’d find out.”
3. Triage the questions into your revision memo (Rep 9) and apply the major fixes to proposal.txt.
4. Commit the revised proposal, the slides, the preregistration link, and the memo to your portfolio with the message midterm: revised proposal after peer review.

If that loop runs clean, you are ready for the midterm. The plan is no longer yours to protect — it’s true enough to defend.

Up next: Project 8