Proposal Presentation + Revised Proposal (MIDTERM)

Apologetic question: "How does iron sharpen iron?"

Project 8 — Proposal Presentation + Revised Written Proposal (MIDTERM)

“The one who states his case first seems right, until the other comes and examines him.” — Proverbs 18:17 (ESV)

Chapter: Chapter 8 — The Proposal, and Defending It Due: End of Week 8 (talk delivered in the Week 8 session; revised proposal due 48 hours after your peer-review session) Weight: 25% of the course — Research Proposal 15% + Proposal Presentation 10%. This is the MIDTERM. Submit: To your Git portfolio and the Canvas dropbox — (1) slides (PDF), (2) the revised proposal.txt (or PDF) in your ACM/IEEE template, (3) your preregistration link, (4) the peer reviews you wrote, (5) your revision memo. Allowed tools: Open. Zotero, Overleaf, the literature tools, and an LLM as a rehearsal partner are all fair game. AI policy: Open-AI for preparing the proposal — but the spine rule holds. Verify every citation an LLM touches against dblp/Semantic Scholar. You stand at the front of the room and answer for every claim; “the model wrote it” is not a defense. Disclose substantive AI use in the proposal’s Acknowledgements, naming the tool and the sections (per your target venue’s policy; see Appendix C). The live talk and Q&A are yours alone — no AI at the podium.

Note on the format. The Litman house midterm is normally a 60-minute closed-AI build. In a research practicum the equivalent live, unaided performance is the proposal defense itself: a 10–15 minute talk plus a Q&A you cannot script, scored in the room. That live defense is the closed-AI core of this midterm; the written proposal is prepared open-AI but defended live. The separate auto-graded methods quiz on Canvas is the closed-book recall check.

The Setup

You have eight weeks of artifacts and one claim worth testing. A funding panel — your committee — will give compute, time, and their attention to exactly the proposals that survive scrutiny. You get 10–15 minutes to convince a room of researchers that your gap is real, your question is falsifiable, your method is sound, and your plan is feasible in the weeks remaining. Then they ask the questions.

This is the most realistic thing you will do all semester. Every grant, every thesis proposal, every conference talk is this same shape: state your case, then have it examined. Proverbs 18:17 is the whole game — the one who states his case first seems right, until the other comes and examines him. Your job this week is to be the kind of researcher who is still right after the examination.

Learning Targets

You will demonstrate that you can:

Assemble a complete, seven-part research proposal from your first-half artifacts.
State a falsifiable claim — RQ, H1, null, and the result that would refute you.
Justify a method another researcher could rerun: named baseline, ablation, splits, seeds, metric, model provenance.
Preregister your analysis plan before any data exists (OSF / AsPredicted).
Deliver a disciplined 10–15 minute talk and defend it live under questions.
Review peers’ proposals specifically and constructively, and revise your own under critique.
Practice research integrity: honest stats plan, verified citations, disclosed AI use, confidentiality of unpublished work.

Normal Tier

The deliverable bar for the midterm.

Required features

Seven-part written proposal in your ACM (acmart, sigconf) or IEEE (IEEEtran) template, started in Week 1: problem & motivation, related work, RQ, method, validity & threats, expected outcomes, timeline.
A falsifiable RQ + hypothesis with the explicit null and refuting result.
A method section naming a non-strawman baseline, at least one ablation, fixed train/val/test splits, a seed/run plan, and an operationalized metric.
A preregistration posted to OSF or AsPredicted, linked in the proposal.
A 10–15 minute talk (≤ 12 slides) delivered live, with a fielded Q&A.
Two peer reviews you wrote, each with a score sheet (out of 35) and one constructive question.
A revision memo triaging the comments you received (major/minor/clarification) and the resulting changes.

Normal-tier rubric (out of 100)

Criterion	Points
Problem & gap is cited, not asserted (points at the matrix)	14
RQ + hypothesis are falsifiable (null + refuting result stated)	14
Method: non-strawman baseline, ablation, splits, seeds, metric	18
Validity & threats section (Cook & Campbell ×4, with mitigations)	10
Preregistration posted and linked	8
Talk delivered within time, ≤12 slides, RQ clear on slide 1	12
Q&A: answered the question asked; honest “I don’t know + how I’d find out”	8
Two peer reviews written: specific, cited, constructive	8
Revision memo: comments triaged, real changes made	8
Total	100

Medium Tier (+ up to 25% extra credit)

Stronger rigor and novelty. Pick what fits your project:

A Registered Report framing: write the proposal so a venue could grant in-principle acceptance on the plan — every analysis decision fixed in advance, no degrees of freedom left open.
A power/sample-size justification for your seed/run plan, not a round number plucked from the air.
Two ablations that isolate two distinct claims, each with the baseline it tests against.
A second baseline from a genuinely different method family (e.g., a classical method and a strong prior LLM approach), so the comparison is not a one-sided fight.
A back-pocket appendix deck that lets you answer at least three anticipated Q&A questions with a prepared figure.

Hard Tier (+ up to 25% additional extra credit)

Publication-ready judgment — the work that makes this a real submission. Required: a memo no tool can write for you.

Write a 2–3 page “Defensibility Memo” addressed to your committee that does what an LLM cannot:

Argue the weakest point of your own proposal and explain why you are proceeding anyway — the tradeoff you’ve chosen, with eyes open.
Name the result you would find most disappointing, and commit now, in writing, to reporting it honestly if it happens. Tie this to your preregistration.
Map your proposal to one concrete target venue (a workshop or Student Research Workshop is the right scale for a practicum draft — see Chapter 16) and state, with evidence, why the contribution fits that venue’s bar.
Make the integrity call: state exactly where you used AI in preparing this proposal, why that use is or isn’t disclosable under your target venue’s policy, and how you verified the citations.

This memo is graded on judgment, honesty, and self-awareness — none of which a model can supply for you, because it requires owning a claim, and ownership is the one thing a tool cannot do.

Submission

Commit to your Git portfolio and submit to Canvas:

portfolio/
├── proposal.pdf            # the revised seven-part proposal (ACM/IEEE template)
├── proposal.docx             # source
├── slides.pdf              # the talk (≤12 slides)
├── prereg-link.txt         # OSF / AsPredicted time-stamped URL
├── reviews/
│   ├── peer-review-1.docx    # a review you wrote
│   └── peer-review-2.docx
├── revision-memo.docx        # triage + changes
└── defensibility-memo.pdf  # Hard tier only

Hints

Build the talk from the proposal, not the other way around. If a slide has no home in the seven sections, cut it.
The empty cell is your slide 4. Photograph or rebuild your comparison matrix; point at the gap.
Rehearse on a timer, out loud, three times (see Rep 5). The third pass reveals the sentence you don’t believe.
Prepare the question you’re most afraid of. Write the answer on a back-pocket slide. The feared question is the gift.
Verify citations the night before, not the morning of. A hallucinated reference in a defended proposal is the worst time to discover one.
For the methods quiz, study from code/midterm-review-map.txt — quickly, like a warm-up. Your real hours go to the proposal.

Cumulative Midterm Review (Weeks 1–7)

The auto-graded methods quiz on Canvas covers the research methods of the first half. It is separate from this proposal; both are due this week. Study the one idea and the one anchor per week:

Wk	Idea to recall	Anchor to name
1	Research = systematic, verifiable new knowledge; the reproducibility crisis	Baker 2016 (~52% see a crisis)
2	A gap is a cited absence (FINER/Heilmeier criteria for a good question)	Swales CARS “establish a niche”
3	Four-layer toolkit (FIND/UNDERSTAND/SNOWBALL/MANAGE); verify every citation	Walters & Wilder 2023 (55%/18% fabricated)
4	Synthesis ≠ summary; the comparison matrix surfaces the gap	rows=papers, cols=method/data/metric/limitation
5	Decision ladder Prompt → RAG → Fine-tune (LoRA/QLoRA) → Distill	retrieval is RAG’s dominant failure mode
6	Falsifiable hypothesis; IV/DV/controls; four validity threats	Cook & Campbell 1979 (statistical conclusion/internal/construct/external)
7	Baselines, ablations, train/val/test, multi-seed, leakage	Bouthillier 2021 (variance); Kapoor & Narayanan 2023 (leakage, 294 papers)

Re-read each chapter’s Why This Matters opener — that is the thesis the quiz tests. Re-do the one rep you found hardest each week; reps are conditioning, the quiz is game day.

What Mastery Looks Like

A master’s proposal is one a stranger could rerun and a skeptic could not easily knock down. The gap is visible in a table, not asserted in a sentence. The hypothesis could come out the other way, and the author has already written down what they’d report if it did. The baseline is the one they’re a little afraid of. And when the hardest question comes in the Q&A, the author doesn’t flinch or bluff — they either answer with evidence or say, plainly, “I don’t know; here’s how I’d find out.” That sentence is the sound of a researcher, not a salesperson.

Coach’s Note — You will be tempted, at the podium, to defend the flawed plan rather than admit the reviewer found the flaw. Resist it. The proposal is not yours to protect; it’s yours to make true. A defense that ends with “you’re right, I’ll fix that” is a win, not a loss.

When You’re Done

A theological footnote. “The one who states his case first seems right, until the other comes and examines him” (Proverbs 18:17, ESV). The verse is a warning against the confidence of the unexamined — and, read with Proverbs 27:17 (“iron sharpens iron”), a gift. We submit our work to examination not because we doubt the truth, but because we doubt ourselves — our blind spots, our wishful readings, our fondness for the result we hoped for. Peer review is built on a confessional honesty: I might be wrong, and I would rather a friend find it now than the world find it later. To state your case and then welcome the one who examines it is not weakness. It is the humility that makes the search for truth trustworthy. The glory of kings is to search things out (Prov 25:2) — and no king searches alone.